Combined Deep Learning Approach for Colorectal Tumor Classification in Endoscopy Images

Colorectal cancer detection in endoscopic images is essential. The Proposed methodology classifies colorectal tumors deeply and reliably by exploiting each method's capabilities. Individual approaches have limitations; thus, the proposed system employs ensemble methods to analyze endoscopic images thoroughly. The classification performance is improved with feature-level and decision-level fusion. Transfer learning from large medical image datasets is used better to adapt the proposed models to colorectal tumor detection. Using the Area Under Curve statistic, the tests show that the proposed system is highly efficient in all metrics compared to the existing systems. The diagnosis accuracy is better than single methods as well. The research advances colorectal cancer detection and illuminates classification technique synergy. The work improves medical image analysis reliability and precision. From 0.88 to 0.95, the proposed method's AUC score, which assesses tumor type recognition, increased. It detects colorectal tumors with 0.90 sensitivity. Specificity was 0.89 for the proposed method. The new method went from 0.90 to 0.96 in accuracy.