Deep Learning Based Detection and Classification of Gastrointestinal Abnormalities from Endoscopic Imaging Data

Gastrointestinal (GI) abnormalities, such as polyps and ulcers, detected through endoscopic imaging are critical for diagnosing severe conditions like colorectal cancer. Accurate detection requires handling challenges such as subtle abnormalities and imbalanced datasets. Traditional detection methods often face difficulties in optimizing model parameters, segmenting abnormalities accurately, and maintaining a balance between speed and precision. The objective is to develop an efficient and robust hybrid technique, Dynamic vortex search-tuned Customized You Only Look Once version 8 (DVS-CYOLOv8), for enhanced detection and classification of GI abnormalities, DVS is utilized for optimizing hyper parameters such as anchor boxes and confidence thresholds while also identifying critical regions of interest. CYOLOv8 leverages advanced segmentation and multi-scale feature detection for real-time performance. A benchmark dataset of annotated endoscopic images covering a range of GI abnormalities is utilized. Preprocessing includes Median Filtering and Contrast Limited Adaptive Histogram Equalization (CLAHE) to suppress noise and improve contrast. Gradient-based techniques are applied for texture and boundary feature extraction. CYOLOv8 captures abnormalities and segments their boundaries with precision, adapting to abnormalities of varying sizes through its multi-scale architecture. DVS optimizes model configuration and enhances sensitivity by focusing on key regions, reducing false positives. It achieves exceptional accuracy (99.11%), precision (98.24%), recall (98.18%), and F1-score (97.66%), outperforming standalone techniques. That presents a scalable and effective clarification for GI abnormality detection, leading to enhanced clinical diagnostics.