FIGURE 4 from Weakly Supervised Deep Learning Predicts Immunotherapy Response in Solid Tumors Based on PD-L1 Expression
Year: 2024
Type: other
Abstract: Explainability for model performance in the pan-cancer-VHIO validation cohort. Overall performance of the model in the validation cohort (A). Distribution of true negative ratio (TNR), true positive ratio (TPR), false negative ratio (FNR), and false positive ratio (FPR) per tumor type in the pan-cancer-VHIO validation cohort (B). Distribution of predicted scores in the pan-cancer-VHIO cohort compared with continuous TPS (C) and CPS (D). Examples of false positives due to histologic differences from NSCLC: PD-L1–stained IHC image from melanoma tumor sample predicted as high PD-L1 by the model (predicted TPS ≥ 1%) with low PD-L1 expression in the tumor cells (TPS < 1%) and high PD-L1 expression in the immune component (CPS ≥ 1%). Magnification of areas with tumor cells with no PD-L1 expression (blue) and immune cells with PD-L1 expression (red), image magnification: 5× and 20× (E). Examples of false negative due to low cellularity samples: PD-L1–stained IHC image from gastrointestinal tumor sample predicted as low PD-L1 by the model (predicted TPS < 1%) with high PD-L1 expression in the tumor cells (TPS = 20%) and high PD-L1 expression in the immune component (CPS ≥ 35%). Magnification of regions with tumor cells, image magnification, 0.35× and 5× (F).
Authors: Marta Ligero, Garazi Serna, Omar S.M. El Nahhas, Irene Sansano, Siarhei Mauchanski, Cristina Viaplana, Julien Calderaro, Rodrigo A. Toledo, Rodrigo Dienstmann, Rami S. Vanguri, Jennifer L. Sauter, Francisco Sanchez-Vega, Sohrab P. Shah, Santiago Ramón y Cajal, Elena Garralda, Paolo Nuciforo, Raquel Perez-Lopez, Jakob Nikolas Kather
Language: English
Subfield: Oncology
Field: Medicine
Domain: Health Sciences
Open Access status: gold
OpenAlex ID: W4390739901