Abstract Introduction: Changes in the number of blood vessels and the proportion of fibrosis regions in xenograft tumors are key indicators for efficacy of oncology drugs, which currently are assessed mainly by human visual examination that is time-consuming, inaccurate and biased because only part of a pathology image is checked. We developed an online AI platform that integrates machine learning algorithms in Computer Vision to quickly and automatically obtain accurate and comprehensive information on blood vessels and fibrosis regions from a whole pathology image. Methods: In a stained pathology image, blood vessels are stained with a different color than other components (e. g. connective tissues, tumor cells, background), so we can directly locate pixels representing them. They also have different topological structures, so we can outline and count the contours of vessels. To count the number of vessels more precisely and reduce the interference of noise, we keep a contour only if it intersects with some located pixels and the number of pixels in the intersection exceeds an adjustable threshold. The fibrosis regions are identified by color contrast since they are stained with a distinct and easily recognizable color. The algorithms were implemented using the OpenCV package. Results: Our AI platform automatically annotates blood vessels and fibrosis regions in pathology images, and generates reports in excel format to report the number and total area of blood vessels, the area and ratio of fibrosis regions, and execution time. It demonstrates superior performance to standard pathology software. Citation Format: Yawen Zheng, Likun Zhang, Dawei Wang, Sheng Gao. An AI platform for vascular and fibrosis analysis of pathology images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 6592.
Read full abstract