AWS Public Sector Blog

Transforming radiology workflows with clinical decision support powered by AWS

Radiology is currently facing some serious global challenges: increasing workloads and complexity of diagnosis, alongside a global radiologist shortage. In 2010, radiologists reviewed 15,000 images a day; by 2018 that number had tripled. In the UK alone, estimates indicate a shortage of 1,939 radiology consultants, leading to a current cost of outsourcing or ad-hoc solutions that has grown to £206 million per year. These factors can lead to delays, missed findings, and high overtime or outsourcing expenses. And data is exacerbating these challenges.

Radiological scans create a significant amount of imaging data. As technology advances, so does the amount of data generated. Radiologists review the available data and report any anomalous findings as quickly as possible. When they encounter a difficult case, they may often remember previous cases—but they typically have no way of retrieving such specific information from large hospital databases. The healthcare technology (HealthTech) startup contextflow uses Amazon Web Services (AWS) to power its clinical decision support system that can help radiologists diagnose patients faster and with more accuracy. The software can search, retrieve, and compare historical diagnostic data to help guide radiologists in writing reports. It is currently being utilized to analyze lung computed tomography (CT) scans for interstitial lung diseases (ILD), chronic obstructive pulmonary disease (COPD), and lung cancer screenings.

Medical image search for large databases

Information search and retrieval is necessary in the practice of radiology, particularly for difficult cases involving ILD and lung cancer, which may require years of specialized training to properly diagnose. When a radiologist is uncertain of a finding, they undergo an information search process to look through text-based resources or consult another radiologist, with a questionable rate of success.

contextflow’s co-founders noted the need to address this issue in 2010 when they first began building a prototype at the Medical University of Vienna. This effort was part of a European research project named KHRESMOI, whose goal was to produce a multilingual, multimodal search-and-retrieval system for large quantities of medical images. They eventually founded their company in 2016 with the goal of empowering radiologists to complete their daily workload faster and with higher quality by delivering software tools that integrate directly into radiologists’ clinical routine. contextflow’s core technology, SEARCH Lung CT, provides radiologists with quantitative and qualitative information for the identification and interpretation of lung disease patterns, COPD, and lung cancer.

contextflow developed SEARCH Lung CT to provide more contextual, objective information for many lung disease patterns, including percentage of lung volume affected by each anomaly, a task that’s incredibly difficult to accurately determine by sight alone. [1] SEARCH Lung CT is a clinical decision support system that searches for 19 disease patterns plus lung nodules in lung CT scans. It provides relevant measurements to add to a radiologist’s structured report and links to differential diagnosis information, all of which are available at the point of care within seconds.

A core component of SEARCH Lung CT [2] is the ability to find cases similar to a current patient’s case. contextflow developed a custom in-house Knowledge Base containing 22,600 lung CTs from 11,600 patients. When a radiologist wishes to further investigate a certain region of a lung CT, they simply draw a bounding box around a region of interest from within their workstation, and SEARCH Lung CT immediately provides similar reference cases, allowing the radiologist to compare and contrast their current patient with thousands of others, boosting confidence and eliminating the need to always wait to speak to a specialist or consult text-based resources.

Indexing and processing with AWS

The clinical insights provided by contexflow SEARCH Lung CT are generated by leveraging multiple deep learning models that provide accurate results within minutes. The models require terabytes (TBs) of data for training and test purposes. Amazon Elastic File System (Amazon EFS) provides a secure and cost-efficient file system to store and retrieve large amounts of data while enabling data sharing between multiple servers, plus other useful built-in features. Amazon EFS Infrequent Access helps contexflow reduce costs by automatically lowering the storage price for files the team rarely uses, while the automatic backup service protects the integrity of data without any additional effort. Sharing data is especially critical, as it enables the scaling and training of multiple model configurations at different Amazon Elastic Compute Cloud (Amazon EC2) instances in parallel. From the compute perspective, Amazon EC2 allows contextflow to use excess servers whenever needed without maintaining servers whenever they are idle. In the production setting, contextflow provides reference results from an in-house Knowledge Base served through Amazon EC2 servers, which helps contextflow provide 24/7 service for healthcare providers.

Figure 1. contextflow SEARCH Lung CT’s user interface provides radiologists qualitative and quantitative information for the interpretation of lung CTs.

Image search for faster radiology reporting

In a 2021 study in collaboration with the Medical University of Vienna (MUW) and Vienna General Hospital (AKH), average radiology report reading time was reduced by 31% when contextflow SEARCH Lung CT was available for use, even though the radiologists searched for additional information more frequently than when the tool was not available [3]. There was also a trend towards improved diagnostic accuracy, and these results held for both junior and senior radiologists. Results are being presented this year at the European Congress of Radiology in Vienna, and the Congress of the European Society of Thoracic Imaging in Oxford. The study’s scientific publication is forthcoming.

For more information about contextflow, its software, and new developments, visit

Are you curious how you can use cloud technology to optimize your workflows in healthcare and more? Reach out to the AWS Public Sector Team for help and more information.

[1] M. Pieler, J. Hofmanninger, R. Donner, A. Sikka, E. Jiménez Arroyo, H. Prosch, R. Zhang, G. Langs, A. Makropoulos. “Evaluation of automatic volumetry of honeycombing and ground glass opacity patterns in lung CT scans.” Presentation at ECR 2022

[2] Röhrich, S., Schlegl, T., Bardach, C., Prosch, H. and Langs, G., 2020. Deep learning detection and quantification of pneumothorax in heterogeneous routine chest computed tomography. European radiology experimental, 4(1), pp.1-11.

[3] S. Röhrich, B. H. Heidinger, G. Langs, M. Krenn, H. Prosch, R. Zhang. “Impact of a content-based image retrieval system on the interpretation of chest CTs of patients with diffuse parenchymal lung disease.” Presentation at ECR 2022.

Read more about AWS for healthcare:

Subscribe to the AWS Public Sector Blog newsletter to get the latest in AWS tools, solutions, and innovations from the public sector delivered to your inbox, or contact us.

Please take a few minutes to share insights regarding your experience with the AWS Public Sector Blog in this survey, and we’ll use feedback from the survey to create more content aligned with the preferences of our readers.

Georg Langs

Georg Langs

Georg Langs is chief scientist at contextflow and professor at the Medical University of Vienna, where he heads the Computational Imaging Research (CIR) Lab. He is a research affiliate at CSAIL, MIT and was work package leader in several EU-funded projects focusing on large-scale medical image retrieval and analysis.

Antonis Makropoulos

Antonis Makropoulos

Antonis Makropoulos is head of research at contextflow, responsible for managing the machine learning team that develops the company's algorithms. Previously he was chief scientific officer at ThinkSono, which develops deep vein thrombosis detection software from ultrasound images. Antonis studied computer science at Athens University of Economics and Business before completing a MSc in artificial intelligence from Edinburgh University and a PhD in medical image processing from Imperial College London.

Razvan Ionasec

Razvan Ionasec

Razvan Ionasec, PhD, MBA, is the technical leader for healthcare at Amazon Web Services in Europe, Middle East, and Africa. His work focuses on helping healthcare customers solve business problems by leveraging technology. Previously, Razvan was the global head of artificial intelligence (AI) products at Siemens Healthineers in charge of AI-Rad Companion, the family of AI-powered and cloud-based digital health solutions for imaging. He holds 30+ patents in AI/ML for medical imaging and has published 70+ international peer-reviewed technical and clinical publications on computer vision, computational modelling, and medical image analysis. Razvan received his PhD in Computer Science from the Technical University Munich and MBA from University of Cambridge, Judge Business School.