HYBRID INTELLIGENCE SYSTEM CNN-LSTM ARCHITECTURE FOR ACCURATE SKIN CANCER CLASSIFICATION USING DERMOSCOPIC IMAGE ANALYSIS

Abstract

Background: Skin cancer, including malignant melanoma and non-melanoma skin lesions, continues to pose a significant global health challenge due to delayed diagnosis and limited access to specialized dermatological expertise. Recent advances in artificial intelligence, particularly deep learning-based dermoscopic image analysis, have demonstrated considerable potential for improving the early detection and diagnosis of skin cancer.

Objectives: This study aims to develop a novel hybrid deep learning framework that integrates Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks to enhance the classification accuracy of dermoscopic skin lesion images by effectively capturing both spatial and contextual feature representations.

Methods: The proposed hybrid CNN-LSTM model was evaluated using the Derm7pt dataset, comprising 8,000 dermoscopic images spanning multiple diagnostic categories. Image preprocessing involved color constancy correction, contrast enhancement, and artifact removal to standardize image quality. Pre-trained CNN architectures, including ResNet50 and EfficientNet-B0, were employed for feature extraction, while LSTM layers modeled sequential dependencies within the extracted feature vectors to capture long-range contextual information. The proposed model was trained and evaluated using standard performance metrics.

Results: Experimental results demonstrate that the hybrid CNN-LSTM framework outperforms standalone CNN models in skin lesion classification. The proposed architecture achieved improved classification accuracy, sensitivity, and F1-score, particularly for minority lesion classes, indicating superior capability in distinguishing visually similar skin lesions while reducing misclassification rates.

Conclusion: The proposed CNN-LSTM architecture effectively combines spatial and sequential feature learning to enhance automated skin cancer classification from dermoscopic images. The improved diagnostic performance highlights its potential for integration into clinical decision support systems, enabling scalable, reliable, and early detection of skin cancer, thereby supporting dermatologists in improving patient outcomes.