Research Article

Automatic Detection of Small Intestinal Hookworms in Capsule Endoscopy Images Based on a Convolutional Neural Network

Table 2

Causes of discrepancies in classifications by the experts and by the CNN.

(%)

False-negative lesions ()
Poorly demarcated(53.6)
 Debris12 (21.4)
 Darkness17 (30.4)
 Poor focus1 (1.8)
Similarity to the edge of the bubble7 (12.5)
Similarity to the submucosal vascular shadow17 (30.4)
Smallness2 (3.6)
False-positive lesions ()
 Darkness32 (3.6)
 Bubble179 (20)
 Debris516 (57.7)
Vascular shadow73 (8.2)
Smallness3 (0.2)
Fold92 (10.3)
Hookworms overlooked by experts ()
 True hookworms2 (100)

Caused by dark view or coffee-like bloody fluid.