Computer Vision News - November 2022

51 Alessa Hering Efficient Tumor Follow-Up Assessment Measurement of metastatic tumors on longitudinal CT scans is essential to evaluate the efficiency of cancer treatment. With yearly increases in the number of imaging studies conducted and the rising global cancer burden, the workload for radiologists keeps increasing. Therefore, there is an urgent need for systems that can speed up reporting and make structured reporting easier. We developed a pipeline that automates the segmentation and measurement of lesions given a point annotation inside the lesion on the baseline scan. Our approach segments the baseline lesion and then uses an image registration method to find the corresponding location in the follow-up image. There, the lesion is also segmented using a CNN. Thereby, the follow-up measurements can be processed fully automatically. In the first step, we evaluated our algorithm with typical metrics like the Dice Score. Even though such metrics give us a rough indication of the performance, it does not tell us how our AI support can impact the follow-up assessment performed by radiologists. Therefore, we conducted a reader study with the goal of comparing a manual workflow with an AI-assisted workflow. In this study, we were able to verify the following three hypotheses: - Assessment time is reduced using the AI- assisted workflow - The inter-reader variability of the segmentation is reduced with AI assistance - AI-assisted segmentation is as good as a manual segmentation I find this research particularly exciting to work on because we work together with radiologists on a specific clinical application.