Fig. 3: HPC performance in mesothelioma subtype classification. | Nature Communications

Fig. 3: HPC performance in mesothelioma subtype classification.

From: A histomorphological atlas of resected mesothelioma discovered by self-supervised learning from 3446 whole-slide images

Fig. 3

a Forest plot for the logistic regression model used in subtype classification, showing log odds ratios (centre) with 95% confidence intervals (error bars), derived from logistic regression coefficients as each HPC's contribution towards subtype prediction, along with significant HPCs for each class (epithelioid versus non-epithelioid), including their p-values, confidence intervals, log odds ratios, and pathologist annotations on HPC histomorphology. This analysis includes n = 3446 WSIs (biological replicates) with subtype labels derived from clinical data. WSIs from epithelioid and non-epithelioid cases (n = 2565 and n = 881, respectively) were treated as independent samples. No technical replicates were included (scale bar, 400 μm). b The ROC (Receiver Operating Characteristic) curve for LATTICe-M test and TCGA-MESO datasets showing their subtype classification performance, including AUC-ROC (Area Under the Curve for ROC), Precision, Sensitivity and Specificity scores. c The PCA (Principal Component Analysis) plot shows patient-level vector representations, color-coded by mesothelioma subtype labels. Source data are provided as a Source Data file.

Back to article page