CF-TRANSNET: COMPACT CONVNEXT-FEMTO_CBAM-TRANSFORMER ARCHITECTURE FOR EARLY DIAGNOSIS OF GASTROINTESTINAL DISORDERS

Abstract

The accurate and timely classification of endoscopic images for gastrointestinal (GI) disorders is vital for early diagnosis. The early diagnosis can help in better patient outcomes. This study focused on development of an AI model for accurate and early diagnosis of GI disorder. This study integrates ConvNeXt’s femto-scale efficient convolutions, Convolutional Block Attention Modules (CBAM), and Transformer encoders to improve spatial–channel feature representation. Kvasir dataset was used in this study for training and testing. The proposed model has achieved an accuracy of 93.37% with a precision of 93.39%, a recall of 93.37%, and an F1-score of 93.37. The model was light weight with only 5.77 million parameters. This work has developed compact yet high-performing architecture for gastrointestinal image classification. This work has shown better results over pretrained baseline models. This work has shown significant gains in spatial and channel attention through optimized attention mechanisms. Experimental comparisons with multiple pretrained models show the efficiency and robustness of the model. This supports the potential of model for real-time clinical decision support in GI disorder detection.