Hybrid Topological Data Analysis and Deep Learning for Basal Cell Carcinoma Diagnosis

Area of Science:

• Dermatology and Computational Intelligence
• Application of Topological Data Analysis in Medical Imaging
• Deep Learning for Cancer Diagnosis

Background:

• Telangiectasia within skin lesions is a key clinical indicator for basal cell carcinoma (BCC).
• Deep learning (DL) models are increasingly used for skin cancer diagnosis and feature analysis.
• Topological Data Analysis (TDA) offers novel methods for extracting information from complex data.

Purpose of the Study:

• To develop a hybrid Topological Data Analysis-Deep Learning (TDA-DL) model for improved basal cell carcinoma (BCC) diagnosis.
• To investigate the utility of topological features extracted from telangiectasia in BCC detection.
• To enhance the performance of automated skin cancer diagnostic systems.

Main Methods:

• A hybrid model integrating TDA and DL was developed using ensemble learning.
• Persistence homology (a TDA technique) was applied to extract topological features from segmented telangiectasia and skin lesions.
• A pre-trained EfficientNet-B5 model was fine-tuned to generate DL features.

Main Results:

• The hybrid TDA-DL model achieved a state-of-the-art diagnostic accuracy of 97.4%.
• The model demonstrated a high Area Under the Curve (AUC) of 0.995 on a test set of 395 skin lesions.
• Analysis confirmed that telangiectasia features significantly improve BCC diagnosis.

Conclusions:

• The integration of TDA with DL enhances the accuracy of basal cell carcinoma diagnosis.
• Topological Data Analysis shows significant potential for improving the performance of deep learning models in medical diagnostics.
• Automated analysis of telangiectasia using TDA-DL offers a promising tool for clinical applications in dermatology.