Advancing clinically meaningful AI algorithms to improve oncologic outcomes

Division Of Basic Sciences - Nci

Investigators

Linked publications & trials

Abstract

Under this project I tackled several critical oncology data issues, specifically a large proportion of time was spent this year on radiation oncology dosimetry data. Two inhouse projects and one collaborative multi-institutional project progressed in 2024-2025 and are ongoing. In house, we were able to surface and structure radiation oncology data spanning 20 years. This manuscript has been submitted for publication to IJROBP. An additional project with the data that was surfaced involved data usage with AI to predict the administration of re-irradiation based on clinical and radiation dosimetry data. In a collaborative project I worked with the American Association of Physicists in Medicine (AAPM) on tracking patient doses in radiation oncology across disparate electronic systems from various vendors. Treatment planning systems (TPS), radiation oncology information systems (ROIS), and electronic health records (EHR) lack uniformity, complicating dose tracking and reporting. To address this, we examined practices in multiple radiation oncology settings and proposed guidelines for current systems. A survey was conducted among members of various professional groups to understand dose reporting practices in TPS, ROIS, and EHR systems. The aim was to identify consistent components and develop guidelines. We identified 6 treatment scenarios where current ROIS defaults fail to accurately represent dose totals. A standardized approach involving 3 reference point types - primary treatment plan reference, dose check, and prescription tracking - was proposed to address these scenarios. This standardization, now published, aligned with the American Association of Physicists in Medicine's Task Group (TG) 263, paves the way for continual development of automated, standardized, interoperable tools, enhancing the ease of sharing reference point information.