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

Table 1

Characteristics of patients in the training and validation datasets.

Characteristics, (%)	Training dataset ()	Validation dataset
		Hookworms ()	Normal ()
No. of images	11236	469	10000
Age (years), mean (SD)
Sex, male	59 (49.6)	17 (42.5)	9 (45.0)
Exam indication for CE
OGIB	94	28	6
Abdominal pain	23	8	14
Anemia	22	7	2
Abnormal results	8	3	1
Abdominal distention	3	1	3
Diarrhea	3	0	1
Screening	3	0	0
Constipation	1	3	0
No. of hookworms
≤3	46	14
>3	73	26
Location of hookworm
Jejunum	69	33
Ileum	13	1
Diffuse	37	6
Eosinophile granulocyte	3	3
Hookworm ovum	3	1
Concomitant lesions
No other lesions	46	16	7
Enteritis	26	8	5
Polyp	19	3	6
Submucosal mass	18	12	4
Angioectasia	13	2	3
Erosion/ulcer	9	7	3
Miscellaneous^	8	3	4

Values are number (%) except where indicated otherwise. SD: standard deviation; CE: wireless capsule endoscopy; OGIB: obscure gastrointestinal bleeding. Hookworm ovum of stool routine. ^The causes of miscellaneous cases included lymphatic dilatation (), diverticulum (), roundworm (), intestinal scar (), and stromal tumor () in training dataset and lymphatic dilatation (), intestinal scar (), vein tumor (n = 1), and stromal tumor () in validation dataset.