From: Value assessment of artificial intelligence in medical imaging: a scoping review
Domain | Topics or themes and outcome measures |
---|---|
(1) The health problem and current use of technology | No details extracted |
(2) Technology aspects | The main topic is interpretability in the sense that we need to avoid the “Black box problem” and the analysis done by the algorithm needs to be transparent to physicians/staff i.e., explainable AI |
Furthermore, risk of bias, possibly causes discrimination issues and validation. The algorithm development method is highlighted, including data quality, the importance of annotation, external evaluation, and reference standards | |
Equipment and IT was a topic mentioning the clinical IT integration and infrastructure | |
OUTCOME MEASURES: interpretability, quality of scans, technical functioning/feasibility | |
(3) Safety assessment | Safety of patients, potential challenges after implementation of AI to the healthcare system |
Reducing side effects and especially radiation dose, data security and protection | |
OUTCOME MEASURES: natural radiation exposure, using clinical knowledge support | |
(4) Clinical effectiveness, e.g. clinical outcomes* | No details extracted |
(5) Economics | The description of the savings and benefits are most often very general, e.g. improved cost-effectiveness |
OUTCOME MEASURES: reduction in workload and time for staff, reduction in the number of biopsies and patients use of medication | |
(6) Ethical analysis | Privacy, consent, obligations, security, awareness of the use of patients’ data, and ownership of the data |
Ethical approval and consider ethical issues of data, algorithms, trained models, and practice | |
Understanding risks vs. benefits, shared/clear decision-making and transparency of results | |
Big questions: “who owns data”, “can data and the algorithm be trusted” and “what is good clinical practice? | |
OUTCOME MEASURES: validity of data, risks versus benefits, patient safety | |
(7) Organisational aspects | Benefits in the form of reductions in workflow and tasks related to imaging for the staff as a result of AI |
The use of additional time related to implementation and training and the challenges related to ensuring acceptability | |
OUTCOME MEASURES: changes in time use for the health care professionals and patient, clinician acceptability measures | |
(8) Patients and social aspects | Patients’ comfortability including easier imaging process and providing access to own data/report in a safe and secured platform |
Better treatment outcome based on the improved clinical decision is the most discussed issue | |
Patients’ satisfaction, as well as clinical benefits, could result in better acceptability of AI technology in the healthcare system | |
OUTCOME MEASURES: the time required for diagnosis, rating for overall satisfaction, help patients make more informed activity choices | |
(9) Legal aspects | Data security and privacy |
Responsibility for misdiagnosis | |
OUTCOME MEASURES: regulatory approvals, consent from patients | |
(10) Development of AI algorithm, performance metrics and validation | No details extracted |
(11) Other aspects | Overpromising language in studies |
Offering the possibility of performing expensive and time-consuming screening programs in countries that otherwise cannot afford them | |
OUTCOME MEASURES: none identified |