Answers are grounded in this paper and cite their sources.
Deep learning has significantly advanced medical image analysis, particularly in the field of endoscopic images. However, these advancements are constrained by the availability of high-quality, annotated medical datasets. This study examines the effect of selective RGB channel jittering as a targeted data augmentation strategy to improve multi-pathological disease detection in endoscopic images. The proposed approach applies channel-specific Gaussian noise to individual RGB channels, implements transfer learning using two different architectures, and evaluates performance across four gastrointestinal conditions: erosion, polyp, tumor, and ulcer from the MedFMC dataset. to prove robustness. The results demonstrate that blue channel jittering consistently improves detection performance by up to 2.7% in accuracy and 3.05% in F1 across all pathologies, while red and green channel jittering significantly degrade performance. This degradation when jittering red and green channels indicates that these channels contain critical discriminative information for gastrointestinal pathology detection, while blue channel enhancement acts as effective regularization. These findings offer important insights for developing targeted data augmentation strategies in medical image analysis.
No takes yet. Share an insight, caveat, or question.
Duygu Çakır (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: