Fig. 3: Performance of the pathomics-based AI platform at the patient level. | npj Precision Oncology

Fig. 3: Performance of the pathomics-based AI platform at the patient level.

From: Tumor cell- and infiltrating immune cell-based supervised learning artificial intelligence multimodal platform for tumor prognosis

Fig. 3

Correlation coefficients and Lasso-Cox regression were utilized for pathomics feature selection through both weakly supervised deep learning (PathWS, a) and supervised deep learning (PathS, e). Kaplan–Meier (KM) analyses were conducted for PathWS (b–d) and PathS (f–h). Based on the median of the expectation of overall survival predicted from the constructed AI platform, oral squamous cell carcinoma patients were stratified into high-risk (overall survival predicted below the median) and low-risk groups (overall survival predicted above the median). i Annotated H&E-stained images, probably maps, and prediction maps of the two AI platforms. Patients in the low-risk group were predominantly assigned a probability label of “0,” whereas those in the high-risk group were mostly assigned a probability label of “1.” H&E, hematoxylin and eosin; c-index, concordance index.

Back to article page