Application of machine learning and deep learning algorithms in osteoporosis classification, diagnosis, prediction and screening.Review

Background and aims: Osteoporosis is one of the most frequent diseases in the elderly, especiallyin postmenopausal women. In addition to the usual methods used to diagnose osteoporosis,Artificial Intelligence (AI) algorithms such as machine learning (ML) and deep learning (DL)have recently found many capabilities for diagnosis, classification, prediction, and screening. Inthis review research, these algorithms have been investigated in the above areas, as well as theircomparison with the usual diagnostic methods and the strengths and weaknesses of each havebeen discussed.Method: For that, articles registered in the last ۵ years in the PubMed database, which dealt withthe use of artificial intelligence in the diagnosis, classification, prediction, and screening, withrelevant terms were analyzed. Terms such as “artificial intelligence”, “deep learning”, “machinelearning”, and “osteoporosis” are searched. The documents were then classified and reviewed,depending on the type of domain in which the AI algorithm was used. The algorithms used in thedesired subjects were examined and the results were presented based on the precision factors ofthe models used.In all, ۴۵ papers were reviewed: ۴ on osteoporosis classification, ۱۴ on osteoporosis prediction, ۱۴on osteoporosis diagnosis, ۷ on osteoporosis screening, and ۶ review papers focused on the use ofthe two artificial intelligence algorithms.Results: Reviewing the studies showed that: in the field of ‘classification’ ۵۰% of studies haveinvestigated the issue using deep learning algorithms and the others using machine learning algorithms.In the field of ‘prediction,’ ۵۷% of studies have investigated the issue using ML, ۳۵.۷%using DL, and ۷.۱% using both ML and DL algorithms. In the field of ‘diagnosis’ ۵۷% of studieshave investigated the issue using ML, ۴۳% using the DL algorithm. In the field of ‘screening’۳۸.۵% of studies have investigated the issue using ML, ۶۲.۵% using the DL algorithm.Dual-energy X-ray absorptiometry (DXA) is underused in the measurement of bone mineraldensity (BMD) and assesses the risk of fracture. Artificial intelligence-based algorithms presentautomated tools for identifying fractures, predicting BMD, and assessing fracture risk throughX-rays that may help identify patients for osteoporosis. Such algorithms can predict BMD by CTscan interpretation and, because of this, can predict the risk of osteoporosis and modified neuralnetworks like CNN can classify and diagnose osteoporosis.Conclusion: While these models are very useful for the classification, diagnosis, prediction, andscreening of osteoporosis, their improvements, such as the inclusion of positive-negative classbias, are maintained. With the ability to expand artificial intelligence algorithms and lower costdevelopment algorithms, it can be expected that these algorithms will widely help specialists inthe near future.

کلیدواژه ها:

artificial intelligence ، deep learning ، machine learning ، X-ray ، bone mineral density ، osteoporosis.