Application of machine learning to determine top predictors of noncalcified coronary burden in psoriasis: An observational cohort study

Machine Learning in Dermatology: Current Applications, Opportunities, and Limitations

“…While coronary plaques can be characterized through coronary computed tomography angiograph, identifying potential risk factors is important for predicting prospective cardiac events. A recent study used ML to identify top predictors of non-calcified coronary burden in psoriasis [98]. These authors identified that obesity, dyslipidemia, and inflammation are important comorbidities/risk factors in atherosclerosis.…”

The Need for a National Strategy on Artificial Intelligence in Canadian Dermatology

“…1 Since then, there has been an explosion of AI-based dermatology research ranging from detection of keratinocyte cancer, 2 prediction of future events such as skin cancer 3 and adverse cutaneous drug reactions, 4 and identification of risk factors for atherosclerosis in psoriasis patients. 5 Indeed, dermatology is ripe for AI technology given our specialty is visually-based with high case volumes for collection of training data. In Canada, this dermatology AI boom has sparked the creation of AI-driven dermatoscopes and medical records systems (MetaOptima Technology Inc.) and noninvasive skin biopsy technology (Elucid Labs Inc).…”

“…In light of the rapidly increasing number of potential applications of AI in dermatology, and the inherent limitations and dangers of inappropriate AI, the American Academy of Dermatology (AAD) released a position statement in 2019 highlighting seven important considerations on AI for dermatologists 13 : (1) model development should include high-quality data that is representative of the population on which the AI is intended, and models should be validated extensively before clinical deployment; (2) clinical deployment of AI should allow for easy integration into clinical workflows, and models should continue to be extensively evaluated, iterated, and monitored in the clinical setting including through clinical trials of efficacy and patient safety, identifying situations where model bias and error can occur; (3) post-marketing surveillance includes continued measurement of outcomes relevant to physicians, patients, and health systems including cost, quality of care, safety, and clinical impact; (4) engagement of physicians and patients to assess their expectations, fears, and knowledge of AI and to help guide development of AI applications toward areas of need and utility; (5) education of patients as to when and how AI will be used in improving their dermatological care, and education of physicians on the limitations of AI and its appropriate uses; (6) privacy and medical-legal issues, such as protecting patient health information during development and deployment of AI applications and in determining responsibility in AI error between the physician and AI manufacturer/distributor; (7) advocacy by physicians and patients to collaborate with policymakers to promote high-quality, clinically useful, and inclusive AI applications.…”

Ranking of a wide multidomain set of predictor variables of children obesity by machine learning variable importance techniques

“…The use of ML to rank predictor variables by their importance has been described for e.g. non-calcified coronary burden 24 , attention-deficit and hyperactivity disorder 25 , and Crohn's disease 26 . In the case of obesity, there is one study where RF has been used to rank variables in the prediction of BMI for adolescent girls 22 , although in that case the set of variables is more restricted both in number and domains, mostly of psychological nature and with no genetic data.…”