HCR.20.076 – Implementing Automatic Detection of Bone Lesions into the Clinical Radiology Workflow

Route: Health care research, sickness prevention and treatment

Cluster question: 105 How can Big Data and technological innovation (e-health) contribute to health care?

This project aims to implement AI techniques into radiology departments in hospitals, focusing on the interaction between AI and medical experts to create a hybrid clinician-AI system. AI-based systems can relieve radiologists from repetitive, time-consuming tasks and detect objects which a radiologist may miss. Hence radiologist or residents can use AI to enhance learning. However, AI systems may result in false positives through misclassifications or false detections and hence should learn from the experts as well. A hybrid clinician-AI system would facilitate a way for experts to override the false internal representations, providing feedback to the AI system. We will focus on the development of such a system to detect and localize osteolytic bone lesions in full body CT scans. A large database of these CT scans is acquired by hospitals in the Netherlands, particularly at the ETZ. Our team comprises of university researchers in medical image analysis and deep learning as well as practicing radiologists. Currently, there are no AI-based systems implemented for bone lesion detection. Automatic lesion detection can reduce error rate and accelerate the clinical workflow, which facilitates diagnosis, sickness prevention and treatment at an earlier stage. Our system would show the radiologists the locations of highly suspicious lesions and incorporate feedback. Secondly, the system should allow the follow up of detected lesions over time in different CT scans of a patient. In this project, we will use state of the art deep learning networks trained on these CT scans and develop a workflow to implement the proposed hybrid system in hospitals. What distinguishes this project is that a large and peripheral hospital is an active participant in its development. It is expected that this project would serve as a guide to any project that requires automated medical image analysis to be implemented into the clinical workflow.

Keywords

Bone Lesions, Deep Learning, Diagnosis of Cancer, Medical Image Analysis, Radiology

Other organisations

Elisabeth-TweeSteden Ziekenhuis

Submitter

Organisation Tilburg University (TiU)
Name Prof. dr. E.O. (Eric) Postma
E-mail E.O.Postma@tilburguniversity.edu
Website https://www.csai.nl