Table 2 (a) Ablation study on the impact of preprocessing.

From: Multi-stage convo-enhanced retinex canny DeepLabv3+ FusionNet for enhanced detection and classification of bleeding regions in GI tract

Metric

Without Preprocessing (mean ± 95% CI)

With Clip-BiRetinexNet (mean ± 95% CI)

Improvement (%)

Paired t-test p-value

Segmentation (mPA)

90.3% ± 1.6%

97.6% ± 0.9%

+8.1%

0.0002

Dice Similarity Coefficient

92.5% ± 1.8%

99.6% ± 0.5%

+7.7%

0.0001

Classification Accuracy

94.1% ± 1.4%

99.2% ± 0.7%

+5.4%

0.0003

Sensitivity

93.2% ± 1.5%

98.5% ± 0.9%

+5.7%

0.0004

Specificity

95.7% ± 1.3%

98.9% ± 0.8%

+3.3%

0.0006

AUC

94.6% ± 1.2%

98.5% ± 0.6%

+4.1%

0.0009

Class

mPA (%)

Dice (%)

Active bleeding

97.9

99.4

Coagulated blood

97.4

99.3

Esophagitis

97.1

99.1

Normal

98.0

99.5

Ulcerative colitis

97.3

99.2

Vascular lesion

98.1

99.6