Early Identification of Skin Cancer Using Region Growing Technique and a Deep Learning Algorithm

Abstract

Skin cancer, comprising both melanoma and non-melanoma forms, is a significant public health concern, constituting approximately 5.9% to 7.8% of annual cancer diagnoses. In Indonesia, the predominant form of the disease is basal cell carcinoma (65.5%), followed by squamous cell carcinoma (23%), and malignant melanoma (7.9%). Several studies have shown that early detection of its melanoma form is essential due to the heightened mortality risk. Therefore, this study aimed to assess the efficacy of the Region Growing + LSTM algorithm in improving detection accuracy compared to LSTM. The novelty of the study lay in addressing the inefficiencies of manual dermoscopy image examination and introducing a novel combination of Region Growing segmentation and Deep Learning LSTM for enhanced detection precision. The results showed that the proposed model could identify segmented areas before classification and achieved 96.62% accuracy, outperforming LSTM's 84%. However, LSTM exhibited shorter training and prediction times (39.3 seconds and 3.2 seconds, respectively) compared to Region Growing + LSTM (17 minutes and 2 seconds for training, 3 minutes and 49 seconds for prediction). Although Region Growing + LSTM offered superior accuracy, it required more time than LSTM, showing potential trade-offs between accuracy and efficiency in skin cancer image detection. Doi: 10.28991/HIJ-2024-05-03-07 Full Text: PDF

Keywords

• lstm, region growing
• corn leaf disease
• public health
• health risk.