Development and validation of an artificial intelligence system for grading colposcopic impressions and guiding biopsies

Table 1 Basic characteristics

Characteristic	No. (%)
Characteristic	Training set	Tuning set	Validation set
Images, total no.	68,037	9751	23,479
Patients, total no.	13,604	1944	3887
Age (years)
24–29	2694 (19.8)	399 (20.5)	988 (25.4)
30–49	8815 (64.8)	1211 (62.3)	2058 (53.0)
50–65	2095 (15.4)	334 (17.2)	841 (21.6)
Referral colposcopy indications^a
Primary screening results^b	9989 (73.4)	1254 (64.5)	2337 (60.1)
Suspicious clinical symptoms^c	3687 (27.1)	735 (37.8)	1650 (42.4)
Distribution of pathological results^d
Normal/benign	4217 (31.0)	591 (30.4)	1205 (31.0)
LSIL	4150 (30.5)	594 (30.6)	1188 (30.5)
HSIL	4489 (33.0)	630 (32.4)	1270 (32.7)
Cancer	748 (5.5)	129 (6.6)	224 (5.8)

Abbreviations: HR-HPV high-risk human papillomavirus, ASC-US atypical squamous cells of undetermined significance, LSIL low-grade squamous intraepithelial lesion, HSIL high-grade squamous intraepithelial lesion
^aThe percentages sum to over 100% due to the overlap of different colposcopy indications
^bPrimary screening results mainly included cytology (HSIL, LSIL, ASC-US, negative) and/or HPV status (HR-HPV positive without genotyping, HPV16/18, other HR-HPV positive, negative)
^cSuspicious clinical symptoms mainly included abnormal genital tract bleeding, suspicious cervical abnormality, unexplained cervicovaginal discharge, and other factors
^dWhen multiple lesions were present in a patient, the highest grade was used as the final pathological diagnosis

ISSN: 1741-7015