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

Figure 3

Representative images classified differently by experts and the CNN in the validation set (green box, region identified as a hookworm by experts; red box, region identified as a hookworm by the convolutional neural network (CNN); number, the probability score of the CNN reading). Samples of false-negative images due to (a) bubble, (b) darkness, (c) debris, and (d) vascular shadow. Samples of false-positive images due to (e) fold, (f) fold, (g) debris, (h) bubble, (i) vascular shadow, and (j) darkness. (k, l) True hookworm detected by the CNN that were missed by the experts.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

(l)