Fig. 4: Interpretability of AI predictions. | npj Digital Medicine

Fig. 4: Interpretability of AI predictions.

From: Noninvasive evaluation and clinical value prediction of tumor-infiltrating neutrophil-to-T-cell ratio in pancreatic ductal adenocarcinoma

Fig. 4: Interpretability of AI predictions.

SHAP analysis of clinical features: importance (a) and effect direction (b). c Grad-CAM highlights key imaging regions. Fusion model (PORCELAIN) SHAP analysis: importance (d) and effect direction (e) of clinical features and DL_Signature. SHAP SHapley Additive exPlanations, DL_Signature Deep Learning Signature, N N-stage, Plt Platelet, LMR Peripheral Blood Lymphocyte-to-Monocyte Ratio, Hb Hemoglobin, NLR Peripheral Blood Neutrophil-to-Lymphocyte Ratio, Glb Globulin, AGR Peripheral Blood Albumin-to-Globulin Ratio, Alb Albumin, mono Monocyte, PLR Peripheral Blood Platelet-to-Lymphocyte Ratio, ALP Alkaline Phosphatase, AST Aspartate Aminotransferase.

Back to article page