Enhanced Multi-Model Deep Learning Framework with Advanced LICU for Improved Skin Disease Detection

Dr.S.SURYA KUMARI1 , MALLARAPU GUNA SINDU2, NAGIREDDY GREESHMA3,
KORLAKUNTA NAVANEETH4, KAMARAPU HARISH5
1Assistant Professor Dept of Information Technology, SV College of Engineering, Tirupati, India.
2B.Tech, Dept of Information Technology, SV College of Engineering, Tirupati, India. 3B.Tech, Dept of
Information Technology, SV College of Engineering, Tirupati, India. 4B.Tech, Dept of Information
Technology, SV College of Engineering, Tirupati, India. 5B.Tech, Dept of Information Technology, SVCollege of Engineering, Tirupati, India.

Abstract-Skin disease diagnosis is a critical area in medical image analysis, demanding early and accurate detection for effective patient treatment. Existing automated systems typically employ a multi-stage deep learning pipeline involving lesion segmentation, coarse classification, and refined classification with the use of Fully-Convolutional Residual Networks (FCRN) and a Lesion Index Calculation Unit (LICU). This existing system achieves high segmentation accuracy and diagnostic precision but suffers from certain limitations like the coarse classification stage tends to generate a significant number of false positives, and the system's adaptability to diverse and rare skin conditions remainslimited. Moreover, the reliance on specific datasets restricts the model’s generalizability, and computational  complexity remains a challenge in real-time clinicaldeployment. To overcome these constraints, the  pro posed future system enhances diagnostic accuracyand clinical usability by incorporating advanced feature selection techniques and dimensionality reduction (PCA and Mutual Information), optimized preprocessing including denoising and color normalization, and integration of more diverse skin disease datasets. This new system further refines lesion characterization by leveraging improved spatial feature analysis and dynamic weighting schemes in the LICU. The benefits include enhanced sensitivity andspecificity with fewer false positives, enabling faster  and more robust processing across diverse skin diseases.This improves diagnostic confidence and supports  timely, precise clinical decisions. Keywords: Fully-Convolutional Residual Networks,Lesion Index Calculation Unit, normalization,diagnostic precision, precise clinical decisions