AN AUTOMATED AND ARTIFICIAL INTELLIGENCE-BASED SYSTEM FOR THE DIAGNOSIS OF SKIN CANCER

Syeda Qanitah Naqvi; Syeda Rabail Zahra; Muhammad Sajid Maqbool; Muqadas Nadeem; Hira Saleem; Mahnoor Zaman

Authors

Syeda Qanitah Naqvi
Syeda Rabail Zahra
Muhammad Sajid Maqbool
Muqadas Nadeem
Hira Saleem
Mahnoor Zaman

Keywords:

Skin Cancer, Deep Learning, Convolutional Neural Network (CNN), Medical Image Analysis, Image Classification, Data Augmentation, Feature Extraction, Early Detection

Abstract

Skin cancer is one of the most prevalent and potentially life-threatening diseases worldwide, where early and accurate diagnosis plays a critical role in improving patient survival rates. This study presents a deep learning-based framework for the automated classification of skin cancer using thermoscopic images.

The proposed methodology consists of a systematic pipeline including image preprocessing, dataset preparation, model development, and performance evaluation. A publicly available dataset comprising 3304 images is utilized, which is divided into training and testing subsets. Various preprocessing techniques such as resizing, normalization, equalization, and data augmentation are applied to enhance image quality, reduce overfitting, and improve model generalization. To perform classification, two advanced Convolutional Neural Network architectures, ResNet101 and ResNet152, are employed due to their ability to learn complex visual features through residual learning. The models are trained using optimized hyperparameters, including appropriate learning rates, batch sizes, and epochs, to achieve efficient convergence and high performance. The experimental results demonstrate that both models achieve outstanding performance, with 100% training accuracy. ResNet101 attains a testing accuracy of 98.35%, while ResNet152 outperforms with a testing accuracy of 99.15%, along with minimal loss values, indicating strong generalization and robustness. A comparative analysis highlights the superior performance of the deeper ResNet152 model in accurately distinguishing between benign and malignant skin lesions. The findings suggest that the proposed framework is highly effective for early detection of skin cancer and has strong potential for integration into computer-aided diagnostic systems. This work contributes to the advancement of automated medical image analysis and supports the development of reliable and efficient tools for clinical decision-making.