Multi-resolution vision transformer model for histopathological skin cancer subtype classification using whole slide...

Digital pathology has significantly advanced cancer diagnosis by enabling high-resolution visualisation and assessment of tissue specimens. However, the manual analysis of these images remains labour-intensive and susceptible to variability, resulting in inconsistencies in diagnosis and treatment decisions. Herein, a multi-resolution model was developed and externally validated for classifying subtypes of skin cancer from whole slide images (WSIs).

A dataset comprising approximately 1.13 million histological patches generated by dividing WSIs from Non-Melanoma Skin Cancer Segmentation (NMSCS) and Heidelberg datasets into non-overlapping tiles. All patches were normalised using the Macenko method prior to training a self-supervised vision transformer (ViT)-based model. The model was designed to classify six classes, including the most common subtypes of skin cancer: basal cell carcinoma (BCC), squamous cell carcinoma (SCC), intraepidermal carcinoma (IEC), Melanoma, Naevi, and Non-cancerous tissue. Multi-resolution data at 10x, 20x, 40x, and 400x magnifications were incorporated to enhance robustness. The model was externally validated on 5,147 slides from 4,066 patients for non-melanoma cancer subtypes. Model performance was evaluated using classification metrics, and the quadratic weighted Cohen's Kappa (κ) score was used to measure the agreement between the model's predictions and the actual labels with a 95 % confidence interval (CI).

The multi-resolution model demonstrated strong classification performance across six classes, achieving an overall κ score of 0.859 (95 % CI: 0.851, 0.866) and 0.898 (95 % CI: 0.892, 0.904) on the validation and testing sets, respectively, reflecting robust performance across diverse skin cancer subtypes. The multi-resolution model for non-melanoma skin cancer exhibited superior performance, achieving an overall κ score of 0.919 (95 % CI: 0.914, 0.924) on the validation set. On the testing set, the κ score ranged from 0.996 to 0.889 with magnifications of 10x, 20x, 40x, and 400x. Attention maps highlighted clinically relevant features for cancerous tissue at different magnifications, aiding interpretability. Additionally, the model obtained a κ score of 0.791 (95 % CI: 0.774, 0.808) on the external data at the slide level, indicating substantial agreement between the model's prediction and the actual label of WSIs.

The multi-resolution model has the potential to assist anatomical pathologists in automatically detecting, highlighting, and classifying subtypes of melanoma and non-melanoma skin cancer subtypes directly from WSIs. Its capability could lead to more effective clinical decision-making in digital pathology, ultimately improving patient outcomes. For non-melanoma skin cancer, the model could be deployed in resource-limited settings with high incidence of the disease and limited access to experienced dermatopathologists.