Table 2 Diagnostic outcomes of included screening examinations
From: Triaging mammography with artificial intelligence: an implementation study
AI result | Radiologist recommendation after screening | Radiologist recommendation after diagnostic workup | Final pathology | |
|---|---|---|---|---|
463 Experimental Group AI-modified workflow (Fig. 2) | 131 Prioritized by AI 28% (131/463) | 33 Additional imaging needed (BIRADS 0) 25% (33/131) | 12 Tissue diagnosis (BIRADS 4 or 5) | 3 Cancer |
9 Benign | ||||
19 Annual Screening (BIRADS 1 or 2) | N/A | |||
2 Short term follow-up (BIRADS 3) | ||||
0 No diagnostic workup | ||||
98 Annual Screening (BIRADS 1 or 2) 75% (98/131) | N/A | 1 Benign (detected on screening MRI) | ||
332 Not Prioritized by AI 72% (332/463) | 46 Additional imaging needed (BIRADS 0) 14% (46/332) | 6 Tissue diagnosis needed (BIRADS 4 or 5) | 0 Cancer | |
6 Benign | ||||
25 Annual Screening (BIRADS 1 or 2) | N/A | |||
9 Short term follow-up (BIRADS 3) | ||||
6 No diagnostic workup | ||||
286 Annual Screening (BIRADS 1 or 2) 86% (286/332) | N/A | |||
392 Control Group Standard workflow (Fig. 2) AI results do not impact participant or radiologist workflow | AI not used to prioritize | 56 Additional imaging needed (BIRADS 0) 14% (56/392) | 15 Tissue diagnosis needed (BIRADS 4 or 5) | 3 Cancer |
12 Benign | ||||
29 Annual Screening (BIRADS 1 or 2) | N/A | |||
9 Short term follow-up (BIRADS 3) | ||||
3 No diagnostic workup | ||||
336 Annual Screening (BIRADS 1 or 2) 86% (336/392) | N/A | 1 Cancer (detected on screening MRI) 1 Benign (detected on screening MRI) |