VisionNet: An efficient vision transformer-based hybrid adaptive networks for eye cancer detection with enhanced cheetah optimizer

• Published in: Biomedical signal processing and control Vol. 97; p. 106673
• Main Authors: B, Akshaya, P, Sakthivel
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.11.2024
• Subjects: Enhanced Cheetah Optimizer; Eye Tumor Detection; MobileNetV2; Residual Attention Network; Vision Transformer-based Hybrid Adaptive Networks; Residual Attention Network; Eye Tumor Detection; Enhanced Cheetah Optimizer; MobileNetV2; Vision Transformer-based Hybrid Adaptive Networks
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2024.106673

•To develop a new ETDM using advanced methods to detect the tumor cells in the eye at an earlier stage.•To propose ViT-HAN for detecting the tumor cells to provide better convergence and faster training performance in less time.•To implement ECO algorithm for optimizing the parameters of RAN and MobileNetV2 models to improve the performance of accuracy, FPR, FNR.•· To compare the developed model with other classifiers and algorithms using various types of performance metrics. Due to the uncontrolled growth of cells, the detection of eye cancer is important in the healthcare sector. Thus, it affects the interior parts of the eye or extra-ocular, affecting the exterior parts of the eye. However, treating the malignant affected area earlier may be advisable in the case of eye cancer. Eye cancer is detected using several deep-learning techniques. These techniques have certain challenges like high computational cost and also require high time for processing. Therefore, it is necessary to overcome these issues using conventional methods for eye cancer detection. Initially, the images for eye cancer detection are taken from internet resources with a total of 140 images. These gathered images are given as input to the eye cancer detection network. Here, the Vision Transformer-based Hybrid Adaptive Networks (ViT-HAN) is used to detect eye cancer with a fused combination of Residual Attention Network (RAN) and MobileNetV2. Further, the RAN and MobileNetV2 parameters are tuned using the Enhanced Cheetah Optimizer (ECO) algorithm. Hence, the developed eye cancer detection model secures an effectively higher performance rate than the conventional eye cancer detection techniques. The accuracy of the proposed model was 97 %, with a false positive rate of 3.0 % and a false negative rate of 3.0 %. Throughout the empirical outcomes, the developed model performs superior to existing techniques. These results emphasize that proposed frameworks could greatly enhance clinical procedures by improving patient outcomes and diagnostic accuracy in ocular oncology.