Towards more efficient tumor follow-up assessment using AI assistance

Measurement of metastatic tumors on longitudinal computer tomography (CT) scans is essential to evaluate the efficacy of cancer treatment. Manual measurements for the diameter-based RECIST (Response Evaluation Criteria In Solid Tumors) criteria are often time-consuming and error-prone. However, those criteria and the execution of the measurements undergo continuous changes. Lesion segmentation assistance based on artificial intelligence (AI) might significantly speed up response evaluation and help to handle the ever-growing mass of image-based staging and follow-up evaluations. Various technical papers investigate the segmentation accuracy of AI algorithms. While these technical measures give a first impression of the performance, they do not yet tell us whether we can add value to the assessment of cancer patients. As a first step to quantify this, the goal of the presented reader study was to compare the workflow of reading follow-up examinations with and without AI assistance to evaluate the impact of the proposed AI-assisted workflow. Our findings support our research hypothesis of an assisted workflow which is superior with respect to processing time and non-inferior with respect to accuracy compared to the manual workflow.