Supplementary Figure S3 from GPSai: A Clinically Validated AI Tool for Tissue of Origin Prediction during Routine Tumor Profiling
Year: 2025
Type: supplementary-materials
Abstract: Calculation of hierarchical positive predictive value (hPPV) and hierarchical sensitivity (hSens). Hierarchical metrics were calculated per sample with values ranging from 0 to 1.0, and then averaged across all samples. The schematic shows level predictions where the “root node” is the broadest category, and not counted toward the calculation. (a) Illustrates 100% hPPV and hSens; (b) illustrates 100% hPPV and 67% hSens; (c) illustrates 67% hPPV and 67% hSens; and (d) illustrates 100% hPPV and 100% hSens, where the model made a more granular prediction than “truth” diagnosis. For example, we commonly receive specimens with a diagnosis of “metastatic carcinoma, consistent with known breast primary” without specifying the subtype. Most likely the cancer was subtyped, but the information wasn’t supplied to us.
Authors: Hassan Ghani, Anthony Helmstetter, Jennifer R. Ribeiro, Todd Maney, Stephanie Rock, Rebecca A. Feldman, Jeff Swensen, Farah Abdulla, David B. Spetzler, Elena Florento, Ari M. Vanderwalde, Patricia Pittman, Milan Radovich, Jaclyn Hechtman, Casey Bales, George W. Sledge, Myra M. George, David Bryant, Jim P. Abraham, Matthew J. Oberley
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4413872672