Table 3.

Summary of study characteristics for studies using AI as reader aid

Study	Study design	Population	Mammography vendor	Index test	Comparator	Reference standard
Pacilè 202030	Enriched test set MRMC laboratory study, counterbalance design (accuracy of a read)	240 women from 1 US centre (50.0% cancer), mean age 59 (range 37-85)	NR	14 MQSA certified radiologists (US) with AI support (MammoScreen version 1, Therapixel, Nice, France); threshold: level of suspicion (0-100) >40	14 MQSA certified radiologists (US) without AI support, single reading; threshold: level of suspicion (0-100) >40	Cancer: histopathology; non-cancer: negative biopsy or negative result at follow-up for ≥18 months
Rodriguez-Ruiz 201932	Enriched test set MRMC laboratory study, fully crossed (accuracy of a read)	240 women (120 from 1 US centre and 120 from 1 German centre; 41.7% cancer), median age 62 (range 39-89)	Hologic 50% Siemens 50%	14 MQSA certified radiologists (US) with AI support (Transpara version 1.3.0, Screenpoint Medical BV, Nijmegen, the Netherlands); threshold: BI-RADS score ≥3	14 MQSA certified radiologists (US) without AI support, single reading; threshold: BI-RADS score ≥3	Cancer: histopathology confirmed cancer; false positives: histopathologic evaluation or negative follow-up for ≥1 year; non-cancer: ≥1 year of negative follow-up
Watanabe 201937	Enriched test set MRMC laboratory study, first without AI support, then AI aided (accuracy of a read)	122 women from 1 US centre (73.8% cancer, all false negative mammograms), mean age 65.4 (range 40-90)	NR	7 MQSA certified radiologists (US) with AI support (cmAssist, CureMetrix, Inc., La Jolla, CA); no threshold	7 MQSA certified radiologists (US) without AI support, single reading; no threshold	Cancer: biopsy proven cancer; non-cancer: BI-RADS 1 and 2 women with a 2 year follow-up of negative diagnosis

AI=artificial intelligence; BI-RADS=Breast Imaging-Reporting and Data System; MQSA=Mammography Quality Standards Act; MRMC=multireader multicase; NR=not reported.