Early Diagnosis and Detection of Skin Cancer Using Deep Neural Network Models and Feature Extraction from Pre-Trained CNNs

Author

Sahil KhannaFollow

Completion Date

Spring 6-13-2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Program or Discipline Name

Analytics

First Advisor

Dr. Katy Valentine

Abstract

Skin cancer is one of the most common and lethal cancer types. While accurate diagnosis at an early stage is essential for skin cancer treatment it remains difficult to achieve in many regions due to lack of sufficient dermatologists and proper diagnostic equipment. Prior studies show Convolutional Neural Network (CNN) models excel at skin lesion classification and consistently achieve better results than standard diagnostic practices. However, the focus of many studies remains confined to image-based learning while neglecting useful patient metadata that could improve prediction accuracy. This research project created a specialized CNN model to classify skin lesions and evaluated whether adding structured patient metadata improves classification results. The study developed a custom CNN with over 20k dermoscopic images from the International Skin Imaging Collaboration (ISIC) which was then compared against pre-trained State of the Art (SOTA) models. Using Light Gradient-Boosting Machine (LightGBM) the top 20 features were selected which were then combined with image features from custom CNN to create a fusion model. The majority of the SOTA models produced lower recall and F1 score values than the custom CNN. Incorporating metadata into the model resulted in higher precision and Area Under the Curve (AUC) scores which demonstrated enhanced detection reliability for malignant cases. The fusion-based model presents an affordable scalable solution for implementing early melanoma detection systems in underprivileged clinical settings.

Recommended Citation

Khanna, S. (2025). Early Diagnosis and Detection of Skin Cancer Using Deep Neural Network Models and Feature Extraction from Pre-Trained CNNs. Retrieved from https://digitalcommons.harrisburgu.edu/dandt/82