Automated Classification of Aberrant p53 Immunohistochemical Patterns in Cutaneous Squamous Cell Carcinoma Using Machine Learning

This project presents a patch-based deep learning approach for classification of P53 immunohistochemical (IHC) patterns in cutaneous squamous cell carcinoma (cSCC). Using a publicly available UCSF (2021) cohort through cBioPortal, TP53 mutation prevalence was analyzed and correlated with corresponding P53 IHC slides. A ResNet-18–based convolutional neural network (CNN) was trained on 224×224 image patches to distinguish normal from abnormal P53 expression patterns. The model achieved high performance at the patch level (AUC 0.9855, accuracy 0.9697), with reduced performance at the slide level due to tissue heterogeneity and other artifacts. The findings demonstrate the feasibility of computational pathology tools for automated IHC pattern recognition and highlight challenges associated with whole-slide variability.