Introduction
Method
-
Focus only on performance,
-
Use of synthetic datasets or data acquired with phantoms,
-
Not conventional clinical imaging of the breast (e.g., thermal imaging, microwave breast imaging),
-
Focus on histopathology images,
-
Conference abstract.
Results
Characteristics of the included publications
Main identified barriers and facilitators of AI in breast clinical imaging
Barriers | Number of paper (%) | Facilitators | Number of paper (%) | ||
---|---|---|---|---|---|
B1 | DATA | F1 | DATA | ||
B1.1 | Data size and variety | 60 (56.1%) | F1.1 | Datasets initiatives | 15 (14%) |
B1.2 | Data quality and data processing | 48 (44.9%) | F1.2 | Algorithmic approaches to address data barriers | 27 (25.2%) |
B1.3 | Data sharing | 15 (14.0%) | F2 | CLINICAL IMPACT | |
B2 | BLACK-BOX AND TRUST | F2.1 | Diagnostic performance | 53 (49.5%) | |
B2.1 | Model transparency | 51 (47.7%) | F2.2 | Clinical workflow | 58 (54.2%) |
B2.2 | Clinician trust | 19 (17.8%) | F3 | ALGORITHMS AND CONCEPTION | |
B2.3 | Patient trust | 11 (10.3%) | F3.1 | Multivariable data | 24 (22.4%) |
B3 | ALGORITHMS AND CONCEPTION | F3.2 | Numerous algorithms | 11 (10.3%) | |
B3.1 | Model architecture | 16 (15.0%) | F4 | EVALUATION AND VALIDATION | |
B3.2 | Technical constraints | 17 (15.9%) | F4.1 | Increased accessibility of AI | 3 (2.8%) |
B3.3 | Multivariable data | 28 (26.2%) | F4.2 | Benchmarking of AI approaches | 6 (5.6%) |
B3.4 | Involvement of stakeholders | 29 (27.1%) | F5 | EDUCATION | |
B4 | EVALUATION AND VALIDATION | F5.1 | AI for education | 5 (4.6%) | |
B4.1 | Meaningful clinical evaluation | 44 (41.1%) | |||
B4.2 | Data variability | 54 (50.5%) | |||
B4.3 | Quality assurance | 17 (15.9%) | |||
B5 | LEGAL, ETHICAL AND ECONOMIC ISSUES | ||||
B5.1 | Liability | 22 (20.6%) | |||
B5.2 | Law and policies | 24 (22.4%) | |||
B5.3 | Fair AI | 22 (20.6%) | |||
B5.4 | Cybersecurity | 10 (9.3%) | |||
B5.5 | Economic issues | 11 (10.3%) | |||
B6 | EDUCATION | ||||
B6.1 | User education | 9 (8.4%) |