Deep Learning-Driven OSCC Recognition: A Multi-Branch Feature Fusion Approach with Optimized Detection and Segmentation
Year: 2026
Type: article
Abstract: This paper presents a Deep Transfer Learning assisted model for automated recognition of Oral Squamous Cell Carcinoma (OSCC) with the use of multi-branch feature fusion model which integrates optimal detection, classification and post segmentation.The suggested scheme initiates with preprocessing of histopathological images data collection and preprocessing are carried by employing noise removal, contrast normalization, and resizing for ensuring uniform input dimensions.After preprocessing the model have feature extraction followed by classification and post segmentation module.Feature extraction is carried using Multi-Residual Dense Swin Transformer (MRDS-Trans) model, which captures discriminative texture and morphological patterns of cancerous tissues effectively.For reducing redundancy and retain most significant features, the Bobcat optimization model (BOA) is used for feature selection.These optimized features improve classifier accuracy and is provided as input for classification scheme which is carried using Deep EfficientNetV2 and MobileNetV2 scheme, by tuning hyperparameters using Energy Valley Optimizer (EVO) for attaining robust convergence and enhanced generalization.For images classified as abnormal, a post-segmentation model is employed using modified YOLO-V8 so as to refine boundaries and thus analyze carcinoma affected areas effectively.At last, explainability is attained over gradient weighted class activation mapping (Grad-CAM) visualization, enabling interpretable insights to model decision.An experimental evaluation illustrates that suggested scheme enhances recognition accuracy, lesion localization precision, and interpretability of diagnosis significantly, thus establishing them as robust tool for computer-aided OSCC analysis and detection.The proposed model performs well in detecting OSCC, with an accuracy of 98.22%, compared to STOA methods.Statistical validation was conducted in order to highlight the effectiveness of the proposed scheme compared to baseline models.
Authors: Sumathy V, C Pretty Diana Cyril
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7172316632
FWCI:
Cites:
Cited by:
Related to: