Disability is a functional limitation with regard to a particular activity and can present in different forms—for example, developmental disabilities (e.g., autism spectrum disorder), disability associated with an injury (e.g., traumatic brain injury), and progressive disabilities (e.g., MS) (2). AI is becoming a beneficial tool to varying extents in the context of disabilities, including visual, cognitive, motor, learning, speech, and language impairments; behavioral problems; and intellectual disabilities. In this section, we discuss the application of AI in disability in general and in a specific case (i.e., disability associated with MS) (11).
In the visual disability area, various studies have investigated visual impairment—for example, identifying the behavioral dynamics of visually impaired infants (26). In the hearing disability area, one study designed and developed an intelligent decision support system that performed audiological investigations to assess degree of hearing loss and suggest appropriate hearing-aid gain value (27). In the realm of cognitive disability, there is research related to the identification and evaluation of cognitive deficits in schizophrenia (28). With regard to motor disabilities, there are several studies related to motor functions, such as the identification of neuroimaging patterns associated with functional impairment in individual patients (29). In the context of learning disabilities, there are studies related to dyslexia—for example, distinguishing between the electroencephalogram signals of normal, low-functioning, and capable dyslexic children when writing words and non-words (30)—and the identification of cognitive profiles in a large heterogeneous sample of struggling learners (31). Regarding speech and language impairments, Themistocleous et al. used AI to identify mild cognitive impairment based on speech (12). In the behavioral problems area, Yassin et al. worked on classification in schizophrenia, autism, and ultra-high-risk and first-episode psychosis (13). Several studies have been performed using AI in the context of autism spectrum disorders, including classification (13), detection (31), diagnosis (32), prediction (33), identification of high-risk cases (23), and identification of factors associated with autism (34). The predictors of visual–motor integration in children with intellectual disability have also been identified using AI (35). Other research areas in disability include prediction of hospital-associated disability (36), prediction of on-road driving ability in healthy older people (37), prediction of swallowing-related quality of life of the elderly living in a local community (38), determination of whether a patient has any geriatric syndromes (39), identification and characterization of cognitive subtypes within the atrial septal defect population (40), identification of cases of Zika-associated birth defects (41), and classification of samples of speech produced by children with developmental disorders versus typically developing children (42, 43).
The existing disability studies have some similarities and differences in terms of the algorithms used. For example, much of the existing research (13, 21, 23, 29, 31, 37, 38) has used support vector machines (SVMs), a supervised ML method. Linear regression (LR), which is also a form of supervised learning, is another of the most widely used methods (13, 14, 23, 33, 34, 35, 36, 41). Other supervised learning algorithms that have been used include random forest (RF) (13, 14, 32, 36, 40, 41, 43) and decision tree (DT) (13, 21, 41) algorithms. Other work has implemented k-nearest neighbors (KNN) (13, 30, and 41), adaptive boosting (13), extreme learning machines (ELMs) (30), gradient-boosted trees (41), linear discriminant analysis (LDA) (42), and natural language processing (NLP) (39, 44).
Deep learning is another branch of ML that enables computers to solve perceptual problems, such as image and speech recognition (45). Some disability research that relies on image and voice data has implemented deep learning algorithms (33). For example, the authors in (12) and (27) implemented artificial neural networks.
Generally speaking, AI applications function through being fed them data, which might be of different categories and types (20). The possible categories of data in this context include demographic, clinical, behavioral, educational, and medication-related data. Possible data types include scalar (the most basic data type), image, text, voice, and unstructured text. In the existing disability research, studies have used demographic (14, 21, 29, 35, 36, 37, 38, 40), clinical (12, 13, 22, 23, 29, 31, 32, 33, 34, 35, 36, 37, 39, 42, 43, 44), behavioral (26, 34, 37, 40), and medication- and procedure-related (43) data. In addition to data of various categories, existing disability research has worked with types of data other than scalar, such as video (26), image (13), and unstructured text (39, 43).
There are several studies and applications of AI in the MS context. For example, in the area of predicting disability progression, Yperman et al. built ML models to predict disability progression in MS after two years (14). They used RF and LR classifiers using evoked potential time series features. Montolío et al. developed ML techniques to diagnose and predict the disability course of MS (18). They developed various classification algorithms—multiple LR, SVM, DT, KNN, naïve Bayes, ensemble classifiers, and a long short-term memory recurrent neural network—using clinical and optical coherence tomography data. The authors in (46) applied unsupervised ML to brain MRI scans acquired in previously published studies, and their findings indicated that this technique was useful in predicting MS disability progression and response to treatment. Pruenza et al. developed a personalized prediction model for the three stages of the disease as a support tool in clinical decision-making for individual MS patients, applying ML and big data techniques (47). Another study predicted Expanded Disability Status Scale (EDSS) scores in MS patients by developing a combined keyword–ML model based on patients’ electronic health record data, including neurology consult notes (44). Similarly, Marzullo et al. built a convolutional neural network model for expanded disability status score estimation based on the brain structural connectivity representation of a MS patient (48). Tommasin et al. evaluated the accuracy of a data-driven approach, such as ML classification, in predicting disability progression in MS (49). Oprea et al. developed a prediction model to estimate disability as measured by the EDSS and outcome probabilities (50). In the context of secondary progressive MS (SPMS), Law et al. built algorithms based on DT, LR, and SVMs to predict SPMS disability progression using EDSS, MS Functional Composite component scores, T2 lesion volume, brain parenchymal fraction, disease duration, age, and gender (21). Another study developed an ML exploration framework for the disease’s evolution to obtain three predictions: one on conversion to the secondary progressive course and two on disease severity with rapid accumulation of disability concerning the sixth and 10th years of progression (51). Lastly, there is a low likelihood of limitations in our research given that it is only a design of a platform, although we may encounter problems in our implementation phase.