Integrating Advanced Neural Network Architectures for Effective Skin Disorder Classification

• Published in: 2024 Asian Conference on Intelligent Technologies (ACOIT) pp. 1 - 5
• Main Authors: Shreya, A, Gautam, G, Dhivyaa, C R
• Format: Conference Proceeding
• Language: English
• Published: IEEE 06.09.2024
• Subjects: Accuracy; Computational modeling; Convolutional Neural Network; Convolutional neural networks; Data Augmentation; Data models; Deep learning; Feature extraction; Long short term memory; Long ShortTerm Memory; Medical diagnostic imaging; Medical services; Skin; Temporal Sequence Processing; Vitiligo Diagnosis
• ISBN: 9798350374933
• DOI: 10.1109/ACOIT62457.2024.10939524

Skin illnesses, such as vitiligo, provide considerable hurdles in proper diagnosis due to their resemblance to other conditions, necessitating time-consuming examinations by medical personnel. Taking advantage of advances in deep learning and medical image processing, this study suggests a novel way to swiftly and accurately classify these disorders. The proposed model uses many convolutional layers to extract detailed features from skin pictures, followed by Long ShortTerm Memory (LSTM) layers to capture temporal dependencies and improve feature representation. This hybrid architecture consists of six convolutional layers that gradually increase in depth and spatial reduction via max-pooling, culminating in two LSTM layers and three fully connected layers for precise classification. By combining CNNs' spatial feature extraction powers with LSTMs' sequence processing skills, our model obtains a high training accuracy of \mathbf{9 3. 4 0 \%} and precision of \mathbf{9 4. 4 8 \%}, proving its usefulness in distinguishing between vitiligo and scar images. This approach shows potential for improving diagnostic accuracy and efficiency in clinical settings, resulting in better patient outcomes and healthcare delivery.