Blackford and contextflow Announce Commercial Partnership to Bring Comprehensive Chest CT Detection Software to Healthcare Providers

Edinburgh, Scotland (November 16th, 2023) – Blackford, the pioneering strategic AI platform and solutions provider, and contextflow today announced a commercial partnership to bring contextflow’s ADVANCE Chest CT solution to healthcare professionals.

Under the partnership, contextflow’s innovative technology will be integrated with Blackford’s advanced enterprise AI platform. Blackford provides healthcare professionals access to an extensive portfolio of medical AI solutions designed to drive clinical efficiency and improve patient outcomes. By integrating contextflow’s advanced detection technology into the Blackford platform, Blackford can offer healthcare providers a powerful tool for detecting ILD, COPD and lung cancer on chest CTs.

“Blackford exists to improve the lives of patients and populations – we can do this by providing tailored AI solutions to healthcare providers around the world to help enhance therapy selection and treatment optimisation”, said Ben Panter, CEO of Blackford. “We are delighted to partner with contextflow to add their advanced tools for lung disease detection, quantification and monitoring of disease progression to our AI portfolio.” 

contextflow’s CE and UKCA marked technology, ADVANCE Chest CT, offers radiologists comprehensive computer-aided detection support for lung cancer, ILD, and COPD patients. The software detects, visualizes and quantifies nodules and lung disease patterns to enhance the speed and quality of radiology reports. Its upcoming malignancy scoring feature has been shown to not only detect lung cancer sooner but also to reduce both false positives and false negatives (*Adams, Scott J et al., JACR September 2022). 

As contextflow’s Chief Commercial Officer Marcel Wassink puts it, “Successful implementation of lung cancer screening programs will require the use of assistive AI to help with earlier detection and manage the workloads. We also understand that lung cancer is only one of many findings relevant to the patient’s wellbeing, and thus we are proud to offer comprehensive support for chest CT that goes beyond cancer to include ILD, COPD, and in the near future, incidental pulmonary embolism. Our partnership with Blackford will accelerate the adoption of this much-needed AI, increasing its accessibility to radiologists and patients alike.”

About Blackford 

Blackford are pioneers in the radiology AI space, with over a decade of experience working in partnership with leading hospitals and ground-breaking technology providers. We operate as a strategic AI partner, providing access to a tried-and-tested core platform, tailored services, and a portfolio of 100+ applications to help healthcare providers unlock the value of AI and improve patient outcomes. 

Our collaboration and recent arms-length acquisition by Bayer ensures that our customers and partners have the support and long-term security needed to underpin successful AI strategies. 

To learn more about Blackford’s tailored approach to AI solutions visit  

About contextflow 

contextflow is a spin-off of the Medical University of Vienna (MUW) and European research project KHRESMOI, supported by the Technical University of Vienna (TU). Founded by a team of AI and engineering experts in July 2016, the company has a strong interest in bringing state-of-the-art machine learning techniques to the market e.g. improved emphysema detection and lung segmentation. Its computer-aided detection software ADVANCE Chest CT is CE Marked and available for clinical use within Europe under the new MDR. Visit for more information.

Other News

Always in the picture thanks to AI

Evangelisches Klinikum Niederrhein introduces contextflow ADVANCE Chest CT in the Pulmonology Department

Lung diseases are among the most common as well as most diverse health problems worldwide. They require precise diagnoses and effective treatment strategies. In this context, artificial intelligence (AI) opens up a promising avenue for detecting a wide variety of disease conditions in affected patients as early as possible. Fanar Othman, head physician at the Clinic for Pulmonary and Bronchial Medicine at Johanniter Hospital Oberhausen, has also discovered the possibilities of deep learning-based technology for himself. He uses the CE certified software contextflow ADVANCE Chest CT to ensure that even when there is a lot to do, the pulmonologist no longer misses anything.

As part of the Lower Rhine Evangelical Hospital and Bethesda Hospital network, the department not only cares for around 3,000 patients a year in its own hospital, but also provides consultative care for three other sites in the western Ruhr region, including Dinslaken and Duisburg. As a former stronghold of the coal and steel industry, the region is also known as the “coal pot”. Particularly among the older generation, the doctors still see many patients whose health was severely affected by their work in the former mines, blast furnaces and factories. As a result, a not insignificant proportion of the medical care provided locally is for occupational lung diseases such as pneumoconiosis, silicosis, asbestosis and, unfortunately, the resulting late effects such as cancer and pleural mesothelioma. 

So there is a lot to do for the total of five pulmonologists under the direction of Fanar Othman. “We have a large catchment area and, simply because of the size of our association, we get to see a large number of rare pathologies in addition to the usual clinical findings,” says the chief physician. “In these cases, it can already become a challenge to make the correct diagnosis and initiate appropriate therapy.” In this context, a single disease may be associated with multiple radiological patterns. Precise characterization is often laborious and examiner-dependent.

Quick access to relevant knowledge

Since last year, contextflow’s AI solution has been providing a remedy for this problem. It takes over the sifting and provision of information from CT examinations that are relevant for diagnostics. The recognition software detects even small changes in the lung parenchyma and relates them to specific diseases. For nodules, it can even show progression. The resulting findings report is generated automatically and is available directly in the PACS viewer within a few minutes. In this way, the tool saves the diagnostician from having to work through thousands of image slices every day, and at the same time, prevents anything important from being overlooked. 

Using an AI system in his department is new territory for Fanar Othman. He was made aware of ADVANCE Chest CT by his physician colleague, the head of radiology, Prof. Dr. Jörg Michael Neuerburg, who has already been using the software successfully for some time. “From the beginning, I liked how affable and user-friendly the application was,” Othman recalls. “You have to take care of practically nothing. The findings generated by the AI are prepared in a simple and clearly understandable way, so you can adopt the results without time-consuming cross-checking. There is not enough time for anything else in daily practice. Everything has to be done in a jiffy.”

Progress controls at the push of a button

When the opportunity arose to take part in a product training course initiated by contextflow, the pulmonologist jumped at the chance. During the online training, he discovered that the software can do a lot more that is valuable for his work: “For example, the AI is able to compare different series of images taken at different points in time. This makes it possible to assess the size development of nodules, which is very important for determining further therapeutics.”

In addition to calculating the diameter in a plane, the system also applies 3D volumetric analysis to evaluate the total mass of a pulmonary nodule. The volumetric measurement method is becoming increasingly important because it offers a more accurate assessment of malignancy risk, or tumor grade, compared to linear measurement, and also helps to better monitor response to therapy – especially when it comes to determining tumor doubling time. This refers to the fact that in most malignant lesions, volume increases first, followed by size.

From the pattern of findings to differential diagnosis

The software solution is also used for texture analysis, e.g. for characterizing parenchymal changes such as ground-glass opacities, reticular pattern or honeycombing. In this context, the distribution of the changes in the lung allows decisive conclusions to be drawn about the underlying clinical picture. However, differential diagnosis is a difficult and complex task due to the sheer number of existing lung pathologies. In addition, it is often the small but subtle differences that matter. Even experienced experts sometimes reach their limits here. “Sometimes it’s difficult to decide with subtle changes in the lower lobe: is this an emphysema bullae or honeycombing? The AI can differentiate such structures incredibly well,” Fanar Othman is pleased to report.

What the Oberhausen head physician also likes is that the structural parenchymal changes can be displayed in percentage form with the help of ADVANCE Chest CT: “When a check-up is due, you can’t always tell with the naked eye whether it has improved after therapy or not. If the response is not that great, it’s helpful to know by what percentage it ultimately got better. A lot of patients ask us explicitly about that, too.” So such clear feedback can have a positive effect on doctor-patient communication.

Overall, the expert is very pleased with how AI has been able to reduce the workload in his department over the past year while increasing diagnostic accuracy. He says his team is also enthusiastic and has already asked him – or rather the AI – for advice in one or two tricky cases. Othman is not worried that a machine could one day outstrip him: “I see it as an opportunity – especially in view of the fact that there is a shortage of staff anyway. AI is therefore a helpful support in the diagnostic process that makes a lot of things easier. In the end, it is still us who bring it all together with the laboratory values, clinical data and samples and come to a decision. That remains the fine art and continues to be the physician’s task.”

Other News

Stets im Bild dank KI

Evangelisches Klinikum Niederrhein führt contextflow ADVANCE Chest CT in der Pneumologie ein

Lungenerkrankungen gehören weltweit zu den am häufigsten auftretenden sowie vielfältigsten Gesundheitsproblemen. Sie erfordern präzise Diagnosen und effektive Behandlungsstrategien. In diesem Zusammenhang eröffnet Künstliche Intelligenz (KI) einen vielversprechenden Weg, um verschiedenste Krankheitsbilder bei betroffenen Patienten so früh wie möglich zu erkennen. Auch Fanar Othman, Chefarzt der Klinik für Lungen- und Bronchialheilkunde am Johanniter Krankenhaus Oberhausen, hat die Möglichkeiten von Deep-Learning-basierter Technologie für sich entdeckt. Im Bereich der computertomographischen Bildgebung nutzt er das CE-zertifizierte KI-Programm contextflow ADVANCE Chest CT. Selbst wenn viel zu tun ist, entgeht dem Pneumologen so nichts mehr.

Als Teil des Verbunds Evangelisches Klinikum Niederrhein und BETHESDA Krankenhaus Duisburg betreut die Abteilung nicht nur jährlich um die 3.000 Patienten im eigenen Haus, sondern übernimmt darüber hinaus noch die konsiliarische Mitbetreuung von drei weiteren Standorten im westlichen Ruhrgebiet, darunter Dinslaken und Duisburg. Als ehemalige Hochburg der Kohle- und Stahlindustrie ist die Region auch als „Kohlenpott“ bekannt. Besonders in der älteren Generation sehen die Ärzte bis heute viele Patienten, denen ihre Arbeit in den früheren Bergwerken, Hochöfen und Fabriken massiv auf die Gesundheit geschlagen ist. Daher entfällt ein nicht unwesentlicher Teil der medizinischen Versorgung vor Ort auf berufsbedingte Lungenerkrankungen wie Staublunge (Pneumokoniose), Silikosen, Asbestosen und leider auch die daraus resultierenden Spätfolgen wie Krebs und Pleuramesotheliome. 

Viel zu tun also für die insgesamt neun Ärzte unter der Leitung von Fanar Othman. „Wir haben ein großes Einzugsgebiet und bekommen allein aufgrund der Größe unseres Verbundes neben den gängigen Krankheitsbildern auch eine Vielzahl seltener Pathologien zu Gesicht, die nicht so alltäglich sind“, so der Chefarzt. „In diesen Fällen kann es schon zur Herausforderung werden, die richtige Diagnose zu stellen und eine entsprechende Therapie einzuleiten.“ Dabei kann eine einzelne Erkrankung mit mehreren radiologischen Mustern verbunden sein. Eine genaue Charakterisierung ist oft mühsam und untersucherabhängig.

Schneller Zugriff auf relevantes Wissen

Seit letztem Jahr schafft die KI-Lösung von contextflow Abhilfe bei diesem Problem. Sie übernimmt das Sichten und Bereitstellen von für die Diagnostik relevanten Informationen aus CT-Untersuchungen. Dabei spürt die Erkennungssoftware selbst kleine Veränderungen im Lungenparenchym auf und stellt sie in Zusammenhang mit bestimmten Erkrankungen und deren Verlauf. Der daraus resultierende Befundbericht wird automatisch generiert und steht innerhalb weniger Minuten direkt im PACS-Viewer zur Verfügung. Auf diese Weise erspart das Tool dem Befunder, sich täglich durch Tausende von Schichtbildaufnahmen zu arbeiten und verhindert gleichzeitig, dass wichtige Erkenntnisse übersehen werden. 

Der Einsatz eines KI-Systems in seiner Abteilung ist Neuland für Fanar Othman. Aufmerksam gemacht wurde er auf ADVANCE Chest CT von seinem Arztkollegen, dem Chefarzt der Radiologie, Prof. Dr. Jörg Michael Neuerburg, der die Software bereits seit längerer Zeit erfolgreich nutzt. „Mir gefiel von Anfang an, wie umgänglich und benutzerfreundlich die Anwendung ist“, erinnert sich Othman. „Man muss sich praktisch um nichts kümmern. Die von der KI erstellten Befunde sind einfach und klar verständlich aufbereitet, sodass man die Ergebnisse ohne aufwendige Gegenprüfung übernehmen kann. Für alles andere bleibt in der täglichen Praxis auch gar nicht die Zeit. Da muss alles zack, zack gehen.“

Verlaufskontrollen auf Knopfdruck

Als sich die Gelegenheit bot, an einer von contextflow initiierten Produktschulung teilzunehmen, zögerte der Pneumologe nicht lang und ergriff die Chance. Während des Onlinetrainings stellte er fest, dass die Software noch einiges mehr kann, was für seine Arbeit von Wert ist: „Zum Beispiel ist die KI in der Lage, unterschiedliche Serien von Aufnahmen, die zu verschiedenen Zeitpunkten gemacht wurden, miteinander zu vergleichen. Dies ermöglicht es, die Größenentwicklung von Rundherden zu beurteilen, was sehr wichtig für das weitere therapeutische Vorgehen ist.“

Neben der Berechnung des Durchmessers in einer Ebene wendet das System auch die 3D-Volumenanalyse an, um die Gesamtmasse einer Lungenläsion zu bewerten. Die volumetrische Messmethode gewinnt zunehmend an Bedeutung, da sie im Vergleich zur linearen Messung eine genauere Einschätzung des Malignitätsrisikos bzw. des Tumorgrads bietet und außerdem dabei hilft, die Reaktion auf eine Therapie besser zu überwachen – insbesondere, wenn es darum geht, die Tumorverdopplungszeit festzustellen. Damit ist gemeint, dass bei den meisten bösartigen Läsionen zuerst das Volumen zunimmt, dann die Größe.

Vom Befundmuster zur Differentialdiagnose

Des Weiteren kommt die Softwarelösung auch bei der Texturanalyse zum Einsatz, z. B. bei der Charakterisierung von Parenchymveränderungen wie Milchglastrübungen, netzartigen retikulären Mustern oder Honigwabenbildung. Dabei lässt die Verteilung der Veränderungen in der Lunge entscheidende Rückschlüsse auf das zugrunde liegende Krankheitsbild zu. Dennoch gestaltet sich die Differentialdiagnose allein aufgrund der schieren Masse an existierenden Lungenpathologien als schwierige und komplexe Aufgabe. Hinzukommt, dass es häufig die kleinen, aber feinen Unterschiede sind, auf die es ankommt. Selbst erfahrene Experten stoßen hier teilweise an ihre Grenzen. „Manchmal ist es schwierig, bei subtilen Veränderungen im Unterlappen zu entscheiden: Ist das ein Emphysembullae oder Honigwabenmuster? Solche Strukturen kann die KI unglaublich gut differenzieren“, freut sich Fanar Othman.

Was dem Oberhausener Chefarzt außerdem gefällt, ist, dass sich die strukturellen Parenchymveränderungen mithilfe von ADVANCE Chest CT in prozentualer Form darstellen lassen: „Wenn eine Kontrolluntersuchung ansteht, kann man nicht immer mit bloßem Auge erkennen, ob es nach der Therapie besser geworden ist oder nicht. Wenn das Ansprechen nicht so großartig ausfällt, ist es hilfreich zu wissen, um wie viel Prozent es letztendlich besser geworden ist. Viele Patienten fragen uns auch explizit danach.“ So ein eindeutiges Feedback kann sich also positiv auf die Arzt-Patienten-Kommunikation auswirken.

Insgesamt zeigt sich der Experte sehr zufrieden damit, wie die KI die Arbeitslast in seiner Abteilung im letzten Jahr reduzieren konnte und gleichzeitig die diagnostische Genauigkeit erhöht hat. Auch sein Team sei begeistert und habe ihn – oder besser gesagt die KI – in dem ein oder anderen kniffeligen Fall schon um Rat gefragt. Dass eine Maschine ihm eines Tages den Rang ablaufen könnte, darüber macht sich Othman keine Sorgen: „Ich empfinde es als Chance – gerade auch vor dem Hintergrund, dass ohnehin Personalmangel herrscht. Die KI stellt daher eine hilfreiche Unterstützung bei der Befundung dar, die vieles erleichtert. Am Ende sind es immer noch wir, die das Ganze mit den Laborwerten, klinischen Daten und Proben zusammenführen und zu einer Entscheidung kommen. Das bleibt die hohe Kunst und weiterhin Aufgabe des Arztes.“

Other News

Improving lung segmentation for higher coverage of clinically-relevant findings: a contextflow whitepaper

This whitepaper was developed by contextflow’s Scientific and R&D teams to explain our approach behind lung segmentation. The full text is listed below. For a pdf copy, click here.


Automatic lung segmentation in computed tomography (CT) is a critical component of computational medical image analysis. While many approaches exist, it remains a challenging problem for patients whose lungs are affected by disease. Here we describe and evaluate a lung segmentation algorithm that yields high accuracy despite disease patterns thanks to a 3D architecture and diverse training data set. We compare our algorithm with the established state-of-the-art algorithm.


Automatic lung segmentation in chest CT scans is relevant for several reasons. It serves as a stepping stone for additional lung-specific analysis such as nodule detection or disease pattern segmentation.  Focusing on the lung area helps to improve execution speed of dependent components and to disambiguate any finding in adjacent areas. By itself, lung segmentation yields a measurement of the lung volumetry, which can be meaningful, and is necessary to assess the proportion of disease patterns covering lung tissue.

The existing computer-assisted solutions for lung segmentation make use of a wide range of approaches with different levels of complexity pertaining to various use cases, such as lung cancer screening, COVID-19, or COPD assessment (Hu et al., 2001; Armato et al., 2004; Sluimer et al., 2005; Mansoor et al., 2015; Shamim et al., 2022).

At contextflow, we segment the lungs (chest cavity) by their most inclusive definition: we include pleural cavity patterns such as pneumothorax and effusion in our segmentation, even though they are technically not part of the lungs, and designate them accordingly, since they are relevant for comprehensive lung reporting. Our algorithms segment many pathological patterns if present, including pleural cavity patterns such as pneumothorax and effusion, using this inclusive lung segmentation as a starting point. In practice, lung segmentation needs to be simultaneously fast (because it is only one step in a host of different processes to analyze the lungs) and robust, as our product focuses on the detection and quantification of lung abnormalities, including comparison with former studies, to automatically detect (major) pathological changes.

A 3D approach towards robust lung segmentation in CT

To achieve robust lung segmentation, we first focused on the construction of an adequate database. We selected cases based on their clinical content, trying to cover as many different lung pathologies and anomalies as possible to ensure maximum diversity and extensive pathologic pattern coverage (see annex for details). We also made sure to cover a wide variety of acquisition protocols, since there is a large technical heterogeneity regarding CT vendors and acquisition parameters from different institutions around the world. In these cases, the lungs were manually annotated by expert radiologists. The annotation process consisted of a loop of annotations and quality checks until the annotation passed the quality check.

Additionally we worked on the choice of a relevant model architecture. We opted for a 3D-UNet-based architecture. One of the advantages of a model using 3D operations is that 3D information can disambiguate findings hard to discern in a 2D slice. For example, it can be hard to differentiate dense pathological patterns like pleural effusion from structures with soft tissue density like the liver on one slice only. In contrast to 2D models, the 3D model is expected to perform better in these cases, as it uses 3D context to make a prediction. Another advantage is that the model produces a segmentation that is consistent axially by design, contrary to a 2D model that processes each slice independently. It was also designed to be faster, finding a compromise between input size, number of parameters and performance.

Comparative evaluation of lung segmentation accuracy

In order to assess the performance of the developed model, we compared it with a state-of-the-art, publicly-available* model (lungmask) for lung segmentation as a baseline (Hofmanninger et al., 2020). The public model is ranked amongst the best on the public challenge LOLA** (LObe and Lung Analysis) and is frequently used for research purposes for organ segmentation. The model adopts a 2D approach for model training.  

We compared the performance between contextflow 2.0 and lungmask on 1722 scans that were sampled from clinical routine without restriction on age, sex, indication or pathology, and that cover findings such as cancer, emphysema, effusion, atelectasis, fibrosis and other pulmonary diseases. For 1694 cases, at least one Region-Of-Interest (ROI) containing a pulmonary pathological pattern is available, while 120 cases have a pixel-wise segmentation mask. All annotations were created by expert radiologists.

Results: segmentation accuracy evaluated by Dice coefficient

We use the widely-adopted Dice coefficient to compare how well the two models’ predictions overlap with an annotation done by a radiologist. The Dice score ranges between 0 (if there is no overlap) and 1 (complete overlap). We report the Dice coefficient for the whole lungs, the right lung and the left lung, measured on the test set in Table 1. We can see that both models achieve very high Dice on average. However, our method tends to achieve higher Dice with a lower standard deviation, which hints to a more robust segmentation performance over the diverse test dataset.

Results: focusing on regions containing disease patterns

To better assess the difference in performance between the two models, we introduce a metric called Region-Of-Interest (ROI) coverage. ROIs are labeled areas that contain a finding. Here, we consider rectangular ROIs on axial slices containing pulmonary pathological patterns. We define ROI coverage as the percentage of ROI centroids that the lung segmentation is able to cover, meaning that the center of the ROI is included in the segmentation. This metric assesses how accurate the lung segmentation model is, despite the presence of pulmonary diseases. This is crucial in its assessment as a tool to support diagnosis and assessment of lung imaging data. 

Annotating ROIs instead of a pixel-wise annotation of pathological patterns is beneficial because it is more time-efficient. In addition, some diffuse lung pathologies are difficult to annotate on a pixel-level even for expert radiologists, so coarsely annotating the entire affected region is more feasible. For this analysis we use ROIs that were created by radiologists on data that cover a wide range of pulmonary pathological patterns. 

* **

We report the ROI coverage figures in Table 2. The model developed by contextflow performs better overall with a coverage of 98.5% of ROIs compared to 96.5% for the open-source model. This effect is especially pronounced for patterns like masses (87.9% vs. 58.6%) and effusion (95.3% vs. 85.6%). This demonstrates that our model covers the lung pathologies more consistently than the open-source model.

Evaluating execution time

In terms of execution time, our model runs faster than the open-source one. We report the numbers for CPU execution on 8 threads for different sizes of CT scans in Table 3. By design, the open-source algorithm’s execution time directly depends on the number of slices in the input CT whereas the new model depends on the FOV (Field Of View) of the input scan. The developed model can generate lung segmentation much faster than the open-source model.


In this article we describe how we developed and evaluated a new lung segmentation algorithm. We compared it to an open-source, state-of-the-art solution over which we showed superiority with a conventional metric (Dice coefficient) and with a new clinical-finding-based metric (ROI coverage), showing a higher robustness to a wide variety of cases. Our developed solution also runs faster, making it scalable in terms of the number of scans we can process. Thus, our solution better meets the requirements for usage in clinical practice.


List of the patterns/pathologies we aim to cover with lung segmentation:

Airway wall thickening, Atelectasis, Bronchiectasis, Bulla, Consolidation, Cyst, Effusion, Emphysema, Ground glass opacification, Honeycombing, Mass, Mosaic attenuation pattern, Nodular pattern, Nodule, Pneumothorax, Pulmonary cavity, Reticular pattern, Tree-in-bud, Fibrosis, Interlobular septal thickening.


Hofmanninger J, Prayer F, Pan J, Röhrich S, Prosch H, Langs G. Automatic lung segmentation in routine imaging is primarily a data diversity problem, not a methodology problem. European Radiology Experimental. 2020 Dec;4(1):1-3.

Mansoor A, Bagci U, Foster B, Xu Z, Papadakis GZ, Folio LR, Udupa JK, Mollura DJ. Segmentation and image analysis of abnormal lungs at CT: current approaches, challenges, and future trends. Radiographics. 2015 Jul;35(4):1056-76.

Shamim S, Awan MJ, Mohd Zain A, Naseem U, Mohammed MA, Garcia-Zapirain B. Automatic COVID-19 lung infection segmentation through modified unet model. Journal of healthcare engineering. 2022 Apr 11;2022.

Armato III SG, Sensakovic WF. Automated lung segmentation for thoracic CT: impact on computer-aided diagnosis1. Academic Radiology. 2004 Sep 1;11(9):1011-21.

Hu S, Hoffman EA, Reinhardt JM. Automatic lung segmentation for accurate quantitation of volumetric X-ray CT images. IEEE transactions on medical imaging. 2001 Jun;20(6):490-8.

Sluimer I, Prokop M, Van Ginneken B. Toward automated segmentation of the pathological lung in CT. IEEE transactions on medical imaging. 2005 Aug 1;24(8):1025-38.

Other News

More than just nodule detection…

Radiologists at the Imapôle Lyon-Villeurbanne group benefit from the support of contextflow AI to detect lung anomalies

Medical imaging, a central pillar of modern medicine, has become indispensable –  a cornerstone of patient diagnosis – and it will be even more so moving forward. Looking towards the future, Imapôle Lyon-Villeurbanne, the medical imaging department of the largest private healthcare establishment in the Lyon region, Médipôle Lyon-Villeurbanne, is fully committed to this belief and has thus integrated contextflow ADVANCE Chest CT into its clinical routine. To better understand the reasons for adopting contextflow, the selection criteria, the deployment experience and the benefits observed, we spoke to Samir Lounis, CEO & General Manager at ImaOne. He manages and directs the Imapôle group’s activities.

Hello Mr. Lounis, could you tell us about Imapôle Lyon-Villeurbanne?

The Imapôle group consists of 10 radiologists responsible for reporting the medical images from two sites: Médipôle Lyon Villeurbanne, Europe’s largest private hospital with over 850 beds, and the Pôle Médical d’OL Vallée in Décines. These two sites carry out more than 800 examinations a day and around 170,000 examinations a year. 

Our team manages this workload, and we are strongly committed to the use of artificial intelligence solutions. We firmly believe in AI’s potential to support our radiologists and transform them into “augmented radiologists” in order to provide more accurate diagnoses and be able to manage a much higher volume of examinations.

What were the motivations and determining factors that led your radiology department to consider adopting contextflow in your clinical practice at Imapôle Lyon-Villeurbanne?

At Imapôle, a significant proportion of our work (∼1/3) is in oncology. This means that we have to interpret a large number of images to monitor or detect pathologies in our patients. In this context, we had been looking for a solution that could help us detect nodules and monitor their evolution in terms of size, particularly in terms of growth or shrinkage. 

Measuring a nodule is a complex task, subject to many variations. It depends on the anatomical plane used for the measurement and the inclination of the nodule itself. In addition, we wanted the measurements to be volumetric and reproducible. With all these factors in mind, we decided to use software. 

We are also involved in a lung cancer screening program at Médipôle – in France, studies are currently being carried out in this field. At Imapôle Lyon Villeurbanne, we have a large pneumology department, and we wanted to offer a solution that was reproducible, efficient and independent of the operator behind the screen. 

Of the various solutions we had identified, contextflow was one of the three finalists, and appeared to be the most efficient and comprehensive, meeting our needs.

What were the selection criteria and what preliminary evaluations were carried out before choosing contextflow’s software for your radiology department?

We explored the market to find solutions suited to our mission, which is to detect and monitor lung nodules over time. We also took into account other criteria such as the time taken to return results. It was essential that the analysis could be carried out quickly, with a return to the PACS and the doctor in around five minutes, in order to maintain our patient management flow. After a comparison of contextflow to other vendors, we chose contextflow because it offers more than just nodule detection and integrates very well into our PACS. 

The other point that was a “game changer” in our choice was contextflow’s ability to look ahead and offer incidental pulmonary embolism detection in the near future. This gives us a tool capable of responding to several of our problems, particularly in oncology, for long-term monitoring, analysis and reproducibility of measurements, as well as analysis and quantification of other pulmonary pathologies like emphysema.

Can you retrace the history of the integration of contextflow’s software in your radiology department, from its implementation to the present day?

contextflow’s technical teams have been extremely responsive. We were able to put them in touch with our IT and PACS teams, and all three teams quickly managed to install the virtual machine to carry out all the tests. We had a fairly tight deadline to achieve a level of integration that would enable us to use the system seamlessly without the doctor having to leave his environment. This was a key factor.

contextflow’s solution is fully integrated into our workflow. Sending is automatic from the modality to the AI solution, and the results are sent back to PACS. So when the doctor reads the examination, they have the contextflow results at their fingertips. The technical support provided by the teams during the start-up phase was extremely responsive, which is very positive for contextflow. The level of integration with PACS is very high.

What were the key stages in the process of implementing contextflow’s software in your radiology department in terms of training, customization and change management?

As far as contextflow is concerned, training took place in two stages. First, there was a preliminary training session which essentially consisted of a product presentation, followed by a second session where the application of the product in clinical routine was presented. We examined a concrete case and analyzed the results obtained. This training was given by videoconference on different dates to suit the availability of the various doctors involved in the project, which was much appreciated. We were able to start using the solution with remote support from both contextflow and our IT and PACS teams. Everything went very smoothly.

After about a month’s use, contextflow offered to accompany our medical teams on site to benefit from their experience. This would also enable us to make personalized adjustments to the use of the product and help discover functionalities that might not have been fully grasped during the initial training sessions.

This support is still ongoing. An application engineer will come next week to meet our teams, and he will also come back if the doctors feel the need. 

Now, as far as contextflow is concerned, the big advantage is that it is not limited to detecting and monitoring nodules over time, which is essential and very important for lung cancer screening, for example, and for monitoring smokers. Rather, it also offers analysis of other pathologies, notably emphysema, which is a very important quantification, especially with a view to the future. 

In the future, it will also enable the detection of incidental pulmonary embolism, a crucial diagnosis in radiology. At Medipôle Lyon Villeurbanne, the largest private emergency department in France, we see around 250 patients a day, about half of whom go through the imaging department, and many of whom benefit from a CT scan. We’re very pleased to be supported by AI-based detection software for these 250 patients because radiologists’ workload is ever-increasing. You end up with 400 to 500 images to analyze per patient. So it’s great to have an artificial intelligence that can accompany you in this detection phase and highlight areas at risk.

That’s why contextflow’s ability to handle new pathologies to be analyzed was a decisive factor in our choice of solution.

How was contextflow’s software integrated into Imapôle’s existing radiology information system to ensure compatibility, interoperability and synchronization of clinical data?

What’s most important is the whole preparatory integration phase. This involves a considerable amount of work over several weeks, during which all the players involved can discuss technical constraints. The end users, in particular the radiologists, can express their expectations and objectives, particularly in terms of how they wish to find the results in his workflow.

The success of this stage is reflected in the fact that, in the end, the radiologist doesn’t need to leave their usual work environment. They open their PACS and work within it – the contextflow results are there without having to open a new program or change windows. The user is not confronted with a totally different interface. In addition, contextflow’s results can be adapted by the radiologist when they disagree, for example, in the case of a false positive nodule. The more transparent we can make the use of contextflow within the PACS and the more successful the integration, and the more the radiologist will use it.

How do you measure the overall satisfaction of contextflow users in your radiology department in terms of user-friendliness, performance and contribution to clinical decision-making?

Every click costs radiologists time and money, so having a well-integrated AI solution with as few clicks as possible was a priority for us. This is a key element in the use of the solution. If we suggest to a radiologist (who is already very busy and subject to the heavy mental workload linked to medical image analysis) additional constraints such as having to navigate between different windows or files, it’s certain that the solution will not be used. They may try it once or twice but will soon realize that it’s time-consuming, and they’ll end up saying to themselves “I’ll do without it” and never come back to it.

If, on the other hand, the whole process is automated, (i.e. the images are acquired by the scanner, automatically sent to contextflow for analysis, and results are sent back to the radiologist in their native working environment) then all they have to do is validate or invalidate the AI’s results for inclusion in the report. The number of clicks is reduced to a minimum. This makes the system extremely user-friendly. What’s more, the degree of integration of the solution with our PACS is extremely advanced, making our dependence on the solution even more beneficial.

What performance indicators and evaluation criteria are used to measure the effectiveness and clinical impact of contextflow’s software in your radiology department?

In terms of our prescribing physicians, we have a large number of pulmonologists and pulmonary oncologists in our department. So we have a team of doctors who specialize in lung diseases. They have been very satisfied with the contextflow software at an advanced level of pulmonary analysis, especially here in Lyon. They have particularly appreciated the ability to detect and track lung nodules over time and to compare results.

When a patient is sent for assessment after three or six months of chemotherapy, it is extremely valuable to have a tool like contextflow to ensure reproducibility of analysis and measurements. This has really been a major asset for our prescribing physicians.

Today, the use of the tool is practically demanded by prescribers, as they have become accustomed to its use. They therefore refer their patients to our center so that their scans can benefit from this additional in-house analysis. As far as our own doctors are concerned, as I mentioned earlier, the more transparent the interface in the workflow, the more it is used. As a result, 100% of lung scans now go through contextflow, benefiting from both medical and AI-assisted dual analysis. 

The feedback we’ve had from talking to doctors clearly shows that the tool has been adopted and used in the same way as the other AI tools we have in our fleet. We have a team of doctors who are forerunners in the adoption of AI, and they are aware of the benefits that artificial intelligence can bring them.

How would you like to see the contextflow solution evolve in the future?

I would love to see contextflow provide a solution for the detection of pulmonary embolism, as this is a real need for all emergency medical imaging departments. This will be of considerable help to emergency physicians and doctors, speeding up patient management and reducing the time lost in analysis. The contextflow team took our feedback seriously and is working towards this.

Still, we are very satisfied with the current solution. contextflow is continuously  improving the specificity and sensitivity of the nodule detection algorithm. Next, they plan to extend the possibilities of thoracic pathology analysis, not only for the lungs, but also for the vessels and heart, as well as for all organs located in the thoracic region. If contextflow could also provide analysis for these elements in the future, that would be a real asset.

AI is seen as the future, but it also raises fears. As a user, you may be both enthusiastic and reticent about certain applications. However, as a human being, you are aware of the implications and limits of AI. It can open the door to a wide range of possibilities. What is your opinion on this subject?

In today’s world, where everything is evolving rapidly – much faster than a human being can adapt – data, both medical and non-medical, is multiplying exponentially. The analysis of this data must also be multiplied.

However, human beings do not have the capacity to adapt instantly to such a flow of data. We may be able to do so in X years’ time, but today, we need solutions that support us in managing this data flow. Sorting and analyzing this data and information is crucial.

As far as I’m concerned, I can say that AI can raise concerns in certain areas. However, I believe that AI will not replace doctors. This is a fact that I have experienced by using these solutions for several years and observing them in our practice.

What is certain, however, is that the doctor who uses AI will replace the doctor who does not. And therein lies the real challenge. The world has evolved faster than human beings can adapt. It therefore needs technological tools. So, the doctor who integrates AI into his practice will surpass the doctor who doesn’t, plain and simple.

Other News

Plus qu’une simple détection de nodules…

Les radiologues du groupe Imapôle Lyon-Villeurbanne bénéficient du support de l’IA de contextflow pour détecter les anomalies pulmonaires.

L’Imagerie Médicale, pilier central de la médecine moderne, est devenue incontournable aujourd’hui. Véritable pierre angulaire du diagnostic patient, elle le sera encore plus demain. Tourné vers le futur, Imapôle Lyon–Villeurbanne, le service d’Imagerie Médicale du plus grand établissement de santé privé de la région lyonnaise, le Médipôle Lyon-Villeurbanne, s’inscrit pleinement dans cette démarche et, dans cette optique, a intégré contextflow ADVANCE Chest CT dans sa routine clinique. Pour mieux comprendre les motivations de l’adoption de contextflow, les critères de sélection, l’expérience de déploiement et les bénéfices observés, nous avons rencontré Samir Lounis, CEO & General Manager chez ImaOne. Il dirige et pilote l’activité du groupe Imapôle.

Bonjour. Pourriez- présenter Imapôle Lyon-Villeurbanne ?

Le groupe Imapôle est composé de 10 radiologues. Ils sont chargés d’interpréter la production d’images médicales de deux sites : Le Médipôle Lyon Villeurbanne, le plus grand hôpital privé d’Europe avec plus de 850 lits, et le Pôle Médical d’OL Vallée à Décines. 

Ces deux sites réalisent plus de 800 examens par jour et environ 170 000 examens par an.

Notre équipe gère cette charge de travail, et nous sommes fortement engagés dans l’utilisation de solutions d’intelligence artificielle.

Nous croyons fermement en leur potentiel pour soutenir nos radiologues et les transformer en “radiologues augmentés” grâce à l’IA, afin d’une part, pouvoir fournir des diagnostics plus précis et d’autre part, pouvoir gérer un volume d’examens beaucoup plus important.

Quelles ont été les motivations et les facteurs déterminants qui ont conduit votre département de radiologie à envisager l’adoption de l’application contextflow dans votre pratique clinique à Imapôle Lyon-Villeurbanne ?

Au sein d’Imapôle, une part importante de notre activité concerne la cancérologie, environ un tiers. Cela implique que nous devons interpréter un grand nombre d’images, en particulier des scanners, pour le suivi ou la détection de pathologies chez nos patients.

Dans ce contexte, nous avons cherché une solution qui puisse nous aider à dépister les lésions et à suivre leur évolution en termes de taille, notamment en ce qui concerne la croissance ou la diminution des lésions. 

Mesurer une lésion est une tâche complexe et sujette à de nombreuses variations. Cela dépend du plan de coupe utilisé pour la mesure et de l’inclinaison de la lésion elle-même, par exemple dans le cas d’une lésion pulmonaire.

De nombreux facteurs entrent en jeu. Nous souhaitions donc que cette mesure puisse  être volumétrique et reproductible.

Partant de tous ces éléments, nous avons décidé d’utiliser un logiciel. Nous sommes également impliqués dans un programme de dépistage du cancer du poumon au sein du Médipôle.

En France, des études sont actuellement menées dans ce domaine. Au sein d’Imapôle Lyon Villeurbanne, nous disposons d’un département de pneumologie important et nous avons voulu proposer une solution reproductible, efficace et indépendante de l’opérateur qui se trouve derrière l’écran.

Parmi les différentes solutions que nous avions identifiées, le logiciel de contextflow, qui figurait parmi les trois finalistes, nous a paru être le plus performant et le plus complet, répondant ainsi à nos besoins.

Quels ont été les critères de sélection et les évaluations préliminaires effectués avant de choisir l’application contextflow pour votre département de radiologie ?

Nous avons exploré le marché pour rechercher des solutions adaptées à notre mission, qui consiste à détecter et suivre les lésions pulmonaires dans le temps, tout en tenant compte d’autres critères tels que le délai de retour des résultats. Il était essentiel que l’analyse puisse être réalisée rapidement, avec un retour dans le PACS et vers le médecin dans un délai de l’ordre d’environ cinq minutes, afin de maintenir notre flux de prise en charge des patients. Après avoir comparé contextflow à d’autres fournisseurs, nous avons choisi contextflow parce qu’il offre plus que la simple détection de nodules et qu’il s’intègre très bien dans notre PACS. 

L’autre point qui a été un « game changer » dans notre choix, c’est la capacité de contextflow à pouvoir se projeter et proposer, dans un avenir proche, la détection des embolies pulmonaires fortuites.

Nous disposons ainsi d’un outil capable de répondre à plusieurs de nos problématiques, notamment en cancérologie, pour le suivi à long terme, l’analyse et la reproductibilité des mesures, ainsi que l’analyse et la quantification d’autres pathologies pulmonaires comme l’emphysème.

Pouvez-vous retracer l’historique de l’intégration de l’application logicielle contextflow dans votre département de radiologie, depuis sa mise en place jusqu’à aujourd’hui ?

Les équipes techniques de contextflow ont été extrêmement réactives. Nous avons pu les mettre en relation avec nos équipes IT et PACS, et les trois équipes ont rapidement réussi à installer la machine virtuelle pour effectuer tous les tests. Nous avions un délai assez court pour atteindre un niveau d’intégration qui nous permettrait une utilisation transparente, sans que le médecin ne quitte son environnement. C’était un élément clé. 

La solution contextflow est entièrement intégrée dans notre flux de travail. Les envois se font automatiquement de la modalité vers la solution d’IA, et les résultats sont renvoyés dans le PACS. Ainsi, lorsque le médecin prend connaissance de l’examen, il dispose des résultats de contextflow. 

L’accompagnement technique et le support des équipes lors du démarrage ont été extrêmement réactifs, ce qui est très positif pour contextflow. Le niveau d’intégration avec notre PACS est très élevé.

Quelles ont été les étapes clés du processus de mise en œuvre de l’application contextflow dans votre département de radiologie en termes de formation, de personnalisation et de gestion du changement ?

En ce qui concerne contextflow, la formation s’est déroulée en deux étapes. Tout d’abord, il y a eu une formation préliminaire qui consistait essentiellement en une présentation du produit, puis une deuxième partie où l’application du produit a été présentée. Nous avons examiné un cas concret et analysé les résultats obtenus. Cette formation a été dispensée en visioconférence à différentes dates, afin de convenir aux disponibilités des différents médecins impliqués dans le projet, ce qui a été très apprécié. 

Nous avons pu démarrer l’utilisation de la solution avec un accompagnement à distance, si nécessaire, tant de la part de l’équipe de contextflow que de notre équipe IT et PACS. Tout s’est très bien passé.

Après environ un mois d’utilisation, contextflow nous a proposé d’accompagner nos équipes médicales sur site, afin de bénéficier de leur expérience. Cela permettrait également d’apporter des ajustements personnalisés à l’utilisation du produit et de leur faire découvrir des fonctionnalités qu’ils n’auraient peut-être pas saisies lors des premières formations.

Cet accompagnement est toujours en cours. Nous aurons un technicien de l’application qui viendra la semaine prochaine pour rencontrer nos équipes. Il pourra également revenir si les médecins en ressentent le besoin. Maintenant, en ce qui concerne contextflow, le gros avantage est qu’ils ne se limitent pas à la détection et au suivi des nodules dans le temps, ce qui est essentiel et très important pour le dépistage du cancer du poumon, par exemple, et le suivi des fumeurs. Mais, au contraire, il permet aussi d’analyser d’autres pathologies, notamment l’emphysème, ce qui est une quantification primordiale, surtout dans une perspective d’avenir.

À l’avenir, il permettra également de détecter les embolies pulmonaires fortuites, un diagnostic crucial en radiologie. Au Medipôle Lyon Villeurbanne, le plus grand service d’urgence privé de France, nous recevons environ 250 patients par jour, dont la moitié environ passe par le service d’imagerie, et beaucoup d’entre eux bénéficient d’un scanner. 

Nous sommes très heureux d’être soutenus par un logiciel de détection basé sur l’IA pour ces 250 patients, car la charge de travail des radiologues ne cesse d’augmenter. On se retrouve avec 400 à 500 images à analyser par patient. C’est donc une bonne chose d’avoir une intelligence artificielle qui peut vous accompagner dans cette phase de détection et mettre en évidence les zones à risque.

C’est pourquoi la capacité de contextflow à prendre en charge de nouvelles pathologies à analyser a également été un facteur déterminant dans le choix de la solution.

Comment l’application contextflow a-t-elle été intégrée dans le système d’information radiologique existant à l’Imapôle, pour assurer la compatibilité, l’interopérabilité et la synchronisation des données cliniques ?

Ce qui est le plus important, c’est toute la phase préparatoire d’intégration. Cela nécessite un travail considérable qui s’étend sur quelques semaines, pendant lesquelles tous les acteurs impliqués peuvent échanger sur les contraintes techniques. L’utilisateur final, notamment le médecin, peut exprimer ses attentes et objectifs, en particulier sur la manière dont il souhaite retrouver les résultats dans son flux de travail.

La réussite de cette étape se traduit par le fait que, finalement, le médecin n’a pas besoin de quitter son environnement de travail habituel. Il ouvre son PACS et y travaille – les résultats contextuels sont là sans avoir à ouvrir un nouveau programme ou à changer de fenêtre. L’utilisateur n’est pas confronté à une interface totalement différente. En outre, les résultats de contextflow peuvent être adaptés par le radiologue en cas de désaccord, par exemple dans le cas d’un nodule faussement positif.

Plus on parvient à apporter de transparence dans l’utilisation de contextflow au sein du PACS, plus l’intégration est réussie et plus le médecin l’utilisera régulièrement.

Comment mesurez-vous la satisfaction globale des utilisateurs de l’application contextflow au sein de votre département de radiologie en termes de convivialité, de performance et de contribution à la prise de décision clinique ?

Chaque clic coûte du temps et de l’argent aux radiologues, c’est pourquoi il était prioritaire pour nous d’avoir une solution d’IA bien intégrée avec le moins de clics possible.

C’est un élément clé dans l’utilisation de la solution.

Si l’on propose à un médecin, qui est déjà très occupé et soumis à une charge mentale importante liée à l’analyse médicale, des contraintes supplémentaires telles que de devoir naviguer entre différentes fenêtres ou dossiers, il est certain que la solution ne sera pas utilisée. Il peut l’essayer une ou deux fois, mais rapidement, il se rendra compte que cela lui prendra du temps et il finira par se dire : “Je vais m’en passer” et il n’y reviendra plus.

En revanche, si l’on automatise l’ensemble du processus, c’est-à-dire que les images sont acquises par le scanner, envoyées automatiquement à l’IA de contextflow pour analyse, que les résultats sont renvoyés au médecin dans son environnement de travail et qu’il n’a plus qu’à valider ou invalider les résultats de l’IA pour les intégrer dans son compte rendu, alors le nombre de clics est réduit au minimum.

Cela permet une convivialité très appréciable. De plus, le degré d’intégration de la solution dans notre PACS est extrêmement poussé, ce qui rend notre dépendance à la solution encore plus bénéfique.

Quels sont les indicateurs de performance et les critères d’évaluation utilisés pour mesurer l’efficacité et l’impact clinique de l’application contextflow dans votre département de radiologie ?

Au niveau de nos prescripteurs, nous avons une grande quantité de pneumologues et de pneumologues-oncologues au sein de notre pôle. Nous avons donc une équipe de médecins spécialisés dans les affections pulmonaires. Ils ont été très satisfaits de l’application contextflow à un niveau avancé de l’analyse pulmonaire, notamment ici à Lyon. Ils ont particulièrement apprécié la capacité de détecter et de suivre les pathologies pulmonaires dans le temps, ainsi que la possibilité de comparer les résultats.

Lorsqu’un patient est envoyé pour une évaluation après trois ou six mois de chimiothérapie, il est extrêmement précieux de disposer d’un outil tel que contextflow pour assurer la reproductibilité de l’analyse et des mesures. Cela a réellement été un atout majeur pour nos médecins prescripteurs.

Aujourd’hui, l’utilisation de l’outil est demandée presque systématiquement par les médecins prescripteurs, car ils se sont habitués à son utilisation. Ils orientent donc leurs patients vers notre centre afin que leurs examens puissent bénéficier de cette analyse complémentaire en interne. En ce qui concerne nos propres médecins, comme je l’ai mentionné précédemment, plus l’interface dans le flux de travail est transparente, plus elle est utilisée.

Ainsi, à l’heure actuelle, 100 % des scanners pulmonaires passent par contextflow, bénéficiant ainsi d’une double analyse à la fois médicale et assistée par IA. 

Les retours que nous avons obtenus en discutant avec les médecins montrent clairement que l’outil a été adopté et utilisé de la même manière que d’autres outils d’IA que nous avons dans notre parc. Nous avons une équipe de médecins précurseurs dans l’adoption de l’IA, et ils sont conscients des avantages que peut leur apporter l’intelligence artificielle.

Comment aimeriez-vous que la solution contextflow évolue à l’avenir ?

J’aimerais beaucoup que contextflow apporte une solution pour la détection d’embolie pulmonaire, car c’est un besoin réel pour tous les services d’imagerie médicale d’urgence. Cela aidera considérablement les urgentistes et les médecins, accélérant ainsi la prise en charge des patients et réduisant le temps perdu lors de l’analyse. L’équipe de contextflow a pris nos remarques au sérieux et travaille dans ce sens.

Nous sommes très satisfaits de la solution actuelle. contextflow améliore continuellement la spécificité et la sensibilité de l’algorithme de détection des nodules. Ensuite, nous envisageons d’étendre les possibilités d’analyse des pathologies thoraciques, pas seulement pour les poumons, mais également pour les vaisseaux et le cœur, ainsi que pour tous les organes situés dans la région thoracique. 

Si à l’avenir, contextflow pouvait également fournir une analyse pour ces éléments, ce serait un véritable atout.

L’IA est considérée comme l’avenir, mais elle suscite également des craintes. En tant qu’utilisateur, vous pouvez être à la fois enthousiaste et réticent vis-à-vis de certaines applications. Cependant, en tant qu’être humain, vous êtes conscient des implications et des limites de l’IA. Cela peut ouvrir la porte à diverses possibilités. Quelle est votre opinion sur ce sujet ?

Dans notre monde où tout évolue rapidement, bien plus rapidement que la capacité d’adaptation d’un être humain, les données, qu’elles soient médicales ou non médicales, sont multipliées de façon exponentielle. L’analyse de ces données doit donc également être multipliée.

Cependant, les êtres humains n’ont pas la capacité d’adaptation instantanée à un tel flux de données. Peut-être serons-nous capables de le faire dans X années, mais aujourd’hui, nous avons besoin de solutions qui nous accompagnent dans la gestion de ce flux de données. Il est crucial de trier et d’analyser ces données et informations.

En ce qui me concerne, je peux dire que l’IA peut susciter des inquiétudes sur certains aspects. Cependant, je pense que l’IA ne remplacera pas les médecins. C’est un fait que j’ai expérimenté en utilisant ces solutions depuis plusieurs années et en les observant dans notre pratique.

En revanche, ce qui est certain, c’est que le médecin qui utilise l’IA remplacera le médecin qui n’utilise pas cette technologie. C’est là que réside le véritable enjeu. Le monde a évolué plus rapidement que la capacité d’adaptation de l’être humain. Il a donc besoin d’outils technologiques. Ainsi, le médecin qui intègre l’IA dans sa pratique surpassera le médecin « tout court ».

Other News

More information and higher diagnostic reliability thanks to AI

Isala overcomes daily challenges with contextflow ADVANCE Chest CT

Health economists predict that the number of patients in Western Europe will double in the next 20 to 25 years. This will of course affect radiology, which will have to prepare for a corresponding increase in the number of examinations and findings. “If the predictions come true, we will have to become more efficient in what we do – as I do not expect the same growth in the number of radiologists. I see artificial intelligence (AI) as part of the possible solution. This applies not only to radiology, but to all activities along the care process,” says Dr. Martijn F. Boomsma, Radiologist at Isala in Zwolle. The Isala hospital group also operates smaller facilities in Meppel, Steenwijk, Kampen and Heerde. The group has a total of 1,200 beds and is one of the largest non-academic teaching hospitals in the Netherlands. The department of Medical Imaging at Isala examines over 1,200 radiological examinations per day.

The Dutch Radiological Society has also defined AI as one of the four most important developments for the next decade, and indeed radiology is already a pioneer in the application of this technology. And that’s a good thing, says Dr. Boomsma: “AI, in combination with an experienced radiologist, increases diagnostic accuracy because it can effectively support and relieve radiologists in certain aspects.” However, clinics and manufacturers would have to prove that the algorithms have a concrete positive impact on health outcomes. This could greatly improve clinical adoption as it would make a clear case for reimbursement for AI and getting AI into national guidelines. Dr. Boomsma believes his discipline is on the right track, even if it still faces a number of hurdles.

Committed partnership at eye level

The department of Medical Imaging has been using diagnostic AI applications for four years and has been using contextflow ADVANCE Chest CT since the end of 2022. Isala runs the solution on its own servers and has integrated it with its image data management system (PACS) from Sectra.

Implementing the AI algorithms was not a plug-and-play process. The hurdle was not only the vendors, but also internal processes. “We had to convince several parties and get the go-ahead from many echelons before we could start. It took a year before the application was really ready for operation, fully integrated, reliable, and with an uptime of 97 percent. The technical integration was then very simple and straightforward,” says Dr. Boomsma. This process requires a lot of commitment and perseverance from everyone involved. But it is also the point at which trust is built. “This is where true partnership shows itself, and we are still experiencing this with contextflow,” the radiologist is pleased to say. 

Dr. Boomsma also notes contextflow’s effort to continuously develop its solution in close cooperation with its users in order to optimize the workflow and thus increase diagnostic value. “This was also a decisive reason for us to choose contextflow. We see a high level of professionalism and agility, as well as the company’s vision to get the best out of AI in thoracic imaging,” he reports. This is also evident in their day-to-day interactions. “Employees respond promptly to inquiries and problems by phone or email. To this end, they always think in terms of solutions and work to solve problems as quickly as possible,” the radiologist elaborates. ADVANCE Chest CT itself stands out because it can be operated without much training. “With a little practice, you can use the software very quickly; it’s user-friendly. But it also changes the way you look at things,” Dr. Boomsma says.

Valuable support for the findings

Dr. Boomsma has developed his own approach to using the AI software. First, he looks at the key images and obtains a global overview. Then he reads the scan, incorporating the referring physician’s specific questions into the findings and preparing his report. “ADVANCE Chest CT helps me identify and quantify nodules. This allows me to clearly identify whether the disease is progressive or stable. Of course, I can also discard or change individual results I don’t consider relevant, as they are false positives or do not relate to the specific question,” says Dr. Boomsma, explaining his workflow. Not only does he use the AI for pulmonary nodules, but also for the detection and quantification of emphysema.

Meanwhile, ADVANCE Chest CT is an important support for reporting at Isala. “On an CT image without contrast, it is very difficult to find a four-millimeter lesion in the perihilar region. But the system reliably shows it to me. It allows me to make sure I haven’t missed anything. It also helps us in our daily work. For example, it is reassuring to know that the AI is always by your side during a nighttime emergency scan and to be sure that everything important can be reported,” says Dr. Boomsma. He also sees the potential to speed up the reporting of findings, citing scans without significant findings as an example. The algorithm can mark these as such, so that the radiologist only has to check them once and can then concentrate on more complex cases.

More than other AI algorithms

“ADVANCE Chest CT gives me much more information than AI solutions from other vendors. It detects and quantifies nodules, emphysema and also fibrosis. The precise information on the extent of manifestation of suspected pathology adds real value to the findings because it can be indicative and correlate with the patient’s condition and  may guide the need for therapy,” emphasizes Dr. Boomsma. Consistency of findings in general remains a challenge, due to different scan settings, inspiration etc. He highlights ADVANCE Chest CT’s SEARCH feature as another unique selling point. With a single mouse click, an overview of similar cases opens from an extensive database, which radiologists can use to support their diagnosis.

The TIMELINE module is a further facilitator. It clearly visualises the changes in the detected nodules over time showing the percentage of growth as well as the time of volume doubling. This is particularly helpful when preparing for tumour boards and multidisciplinary meetings. The integration with the Sectra PACS gives the potential to seamlessly incorporate the results into the report.

Other News

AI Interest Group for Imaging (AIGI) Task Force

The “Interest Group for AI in Imaging” (AIGI) started its work in January and advocates the optimization and greater use of IHE profiles when using AI (Artificial Intelligence). 

AI applications are slowly becoming the standard in radiological reporting, but there’s a lack of standardization when it comes to the implementation of AI. Two IHE (Integrating the Healthcare Enterprise) integration profiles already exist (IHE AIW-I & IHE AIR) for the integration of artificial intelligence (AI) and its communication with DICOM data, but so far there are only very few implementations in current products visible. Thus, Marc Kämmerer, member of the IHE Europe Steering Committee and Head of Innovation Management at VISUS Health IT GmbH, initiated AIGI.

What is AIGI?

AIGI is a task force of IHE Europe consisting of radiology AI users, software and PACS vendors, marketplace operators as well as other interest groups. Its main interest and goal is to define the means for a standardized and applicable data workflow for actual use cases in European healthcare systems, including how to:

  • deploy and maintain AI applications
  • connect AI applications with end users’ systems
  • integrate the AI application output in end-users’ systems
  • collect and provide end user feedback

Examining the entire workflow between user and AI

The demand for standardization is high from all sides of the workflow equation. AIGI is examining the entire process chain between the user and the AI solution for practical applicability and feasibility on the basis of the profiles mentioned. Looking for holes in the existing standards, the group will create proposals and frameworks for improved, bi-directional data flow between AI and the user.

One initial area of focus: currently when a radiologist sends DICOM data to an AI software, there is no feedback as to how long it will take for the result to reach the PACS. Should the radiologist wait or start with the next patient? A status query could be integrated as standard via the IHE AIW-I profile, and that is exactly the type of question AIGI hopes to standardize.

The same applies to error messages, which previously only took the form of empty reports or system crashes. Similar to DICOM email, feedback should be given here as to whether and what type of error has occurred. In terms of interoperability, the content of these messages would have to be defined so that they have the same meaningfulness independent of the display system. 

Another example: When reporting, it can make sense to play certain evaluations prominently on the surface – for example, degrees of malignancy in mammography. In principle, this is possible via DIOCM-SR objects, but here too the challenges lie in the details.

Simplified AI access for all

“Working on these standards is very important in order to implement binding structures in the AI ​​processes as early as possible. We are currently noticing that many AI providers and AI marketplace operators are working with APIs. However, it is impossible for PACS manufacturers to use all the APIs available on the market. There are currently well over 20 AI marketplace operators and several hundred AI providers. Here we have to find solutions quickly that ultimately benefit everyone involved – manufacturers and users. From the positive response to our task force, we can see that fortunately all parties involved see things the same way. And it didn’t take long for the IHE to convince the group to set up the group,” says Marc Kämmerer, pleased with the response from industry and practice.

contextflow is a proud supporter of AIGI. Currently we are participating in the subgroup related to Longitudinal Data, Reporting, and Pseudo-/Anonymization. The AIGI Taskforce is open to additional members. Its intended output are best practice white papers, correction proposals, and work item proposals to ultimately benefit users, PACS and AI manufacturers and AI marketplace operators. For more information, click here.

Other News

contextflow approved for EIC Accelerator Equity Investment

We are beyond thrilled to announce that contextflow’s application for the European Innovation Council (EIC) Accelerator equity investment program has been accepted!!!

The latest round of EIC funding saw stiff competition: 139 companies were interviewed by juries of experienced investors and entrepreneurs, out of a total of 551 full proposals submitted. We are the only Austrian company selected in the latest round of funding.

contextflow’s history with the EIC dates back to 2020, when we became the fortunate recipients of a €1.2 Million EIC grant. This shows a clear commitment from the European Commission to support European-born companies with cutting-edge innovation over the long-term. The trust bestowed upon us does not go unnoticed; the best way to show gratitude is to continue to develop our comprehensive clinical decision support for chest CT to enable radiologists to perform their routine tasks faster and with higher certainty and accuracy.

An additional HUGE thanks our early supporters and stakeholders for taking us from university spinoff to the 40+ member team we are today.

Technische Universität Wien
Medizinische Universität Wien
TU Wien Innovation Incubation Center (i²c)
INiTS | Vienna’s High-Tech Incubator
Health Hub Vienna
LISAvienna – Life Science Austria Vienna
FFG Österreichische Forschungsförderungsgesellschaft mbH
Vienna Business Agency
Austria Wirtschaftsservice

Special personal thanks to Angelo Nuzzo, PhD, MBA, MA, Birgit Hofreiter, Alexandra Negoescu & Irene Fialka for guiding us from the very beginning. We couldn’t have done it without your support!

Other News

Radiology Symposium Mainz returns for a 2nd year

An impressive crowd at this year’s Radiology Symposium Mainz! For a second year running, we gathered at the Bootshaus Mainz at the end of June to discuss current radiology AI topics.

Thank you to our fabulous radiology speakers & partners, starting with host Prof. Dr. med. Peter Mildenberger from University Medical Center Mainz. Univ.-Prof. Dr. med. Christoph Düber, Prof. Dr. med. Mike Notohamiprodjo, Elodie Weber, Erwin Krikken, Karin Klein, Marcel Wassink, Mark Rawanschad, PD Dr. Christian Elsner, Florian Brandt, Knut Dietrich-Thiel, Bernd Schütze, Alex Lemm, Barbara Lampl, PD Dr. Daniel Pinto dos Santos, Prof. Dr. med. Elmar Kotter, medavis GmbH, EIZO Healthcare, DFC-SYSTEMS GmbH.

We would also like to thank our friends at Sectra for their continued collaboration!

Hope to see you again next year!

Other News