Enhanced Gastrointestinal Disease Detection Using Resnet50: A Comparative Study Of Deep Learning Architectures For Medical Image Classification

Abstract

Gastrointestinal (GI) Diseases Such As Polyps, Ulcers, And Tumors Require Early Detection For Effective Treatment. Traditional Manual Inspection Of Endoscopy Images Is Timeconsuming And Prone To Human Error. This Project Proposes An AIbased System Using ResNet50 For Automatic Multiclass Classification (Normal, Polyp, Ulcer, Tumor) Of GI Diseases From Endoscopy Images. A Comprehensive Comparative Analysis Of VGG16, MobileNetV2, EfficientNetB0, And ResNet50 Is Conducted On The Kvasir Dataset (8,000 Images). ResNet50 Achieves The Highest Performance: 94.2% Accuracy, 93.8% Precision, 94.1% Recall, And 93.9% F1score, Outperforming VGG16 (89.4%), MobileNetV2 (86.7%), And EfficientNetB0 (91.3%). The System Reduces Diagnosis Time From 8–10 Minutes Per Patient To Under 2 Seconds Per Image, Improving Diagnostic Consistency And Early Detection.

Keywords