Machine Learning for Stroke Rehabilitation in Low- and Middle Income Countries

Even in developed countries stroke rehab is a failed shitshow. Don't try to emulate them, they know nothing, just go straight to creating 100% recovery protocols. 

Machine Learning for Stroke Rehabilitation in Low- and Middle Income Countries

Chawla N. S.1 , Rana S. 2 1Department of Neurology, Christian Medical College and Hospital, Ludhiana, Punjab (141008). 2School of Electronics & Electrical Engineering, Lovely Professional University, Phagwara, Punjab (144001). 

Abstract 

Stroke rehabilitation plays an integral role in the path of recovery. In developed countries, the rehabilitation delivery is well equipped and access to these services is easy for the majority. However, in low and middle income countries (LMICs), there are a lot of barriers to implementation of effective rehabilitation services post stroke. Integrating technology into rehabilitation would facilitate this process. Various machine learning algorithms have been shown to ameliorate the outcome measurement and rehabilitation delivery process. This review aims to highlight the available literature from LMICs in stroke rehabilitation and throw some light on the gaps of available research data. Keywords Stroke rehabilitation, stroke, neurological rehabilitation, technology-based rehabilitation, machine learning, deep learning, neural network, feature engineering 

Introduction 

Stroke is one of the main culprits of death and disability across the globe. 1 Its incidence in reported in exponentially higher numbers in lowand middle income countries (LMICs) as compared to high income countries(HICs).1 The extensive disease burden is implicated for the morbidity and mortality rates in LMICs.2,3 It has been well established that rehabilitation following stroke positively impacts the prognosis of the patient and substantially reduces the impairments and handicap.4 Stroke can greatly impact the quality of life of the affected person by causing activity limitation and participation restriction. The aim of stroke rehabilitation is to reduce his limitation and restriction and make the patient as independent as possible. However, the status of rehabilitation differs immensely in HICs and LMICs.4 In HICs, there are set clinical practice guidelines to give direction to the stroke rehabilitation process. There is a proper infrastructure for rehabilitation and easy access is ensured to the majority of population.5 The higher socioeconomic status and good educational level also facilitates proper delivery of rehabilitation to stroke patients in HICs.5 However, in contrast to HICs, LMICs face a lot of barriers to the delivery of stroke rehabilitation services. Lack of awareness regarding rehabilitation is one of the major constraints in addition to socio-economic factors.4There is also a generalised lack of healthcare infrastructure. The rehabilitation centres are under-sourced. The low therapist to patient ratio © 2022 IJRAR May 2022, Volume 9, Issue 2 www.ijrar.org (E-ISSN 2348-1269, P- ISSN 2349-5138) IJRAR22B2384 International Journal of Research and Analytical Reviews (IJRAR) www.ijrar.org 423 only adds to burnout and further compromises the quality of care provided by rehabilitation professionals.4 It is clearly warranted that concrete strategies be set in place to improve the status of stroke rehabilitation delivery in LMICs. Thus an appropriate solution needs to be devised to counter these barriers and provide effective stroke rehabilitation in LMICs. One of these strategies is the integration of cost-effective technology into rehabilitation.6 With advances in technology, machine learning has emerged as a boon in the field of stroke rehabilitation.6 Therefore the main objective of this review is to explore the recent advances of machine learning in healthcare and highlighting some machine learning based algorithms for stroke rehabilitation in LMICs. Methodology Published scientific literature related to machine learning and healthcare in lower and middle income countries was collected from online databases. The databases selected for this search were Pubmed and Google Scholar. Some relevant keywords used to carry out this search were identified. These included machine learning, deep learning, neural network, artificial intelligence, healthcare, health, rehabilitation, stroke, cardiovascular diseases, cardiac diseases, low and middle income countries, developing countries, etc. The studies thus retrieved were screened based on whether or not they fit into the objectives of this review. Data was extracted and represented in a meaningful manner. Discussion Machine learning is a field of study that allows a computer to learn and function like a human brain. It uses data and experience to improve accuracy.7 There are three ways to approach a classical machine learning algorithm- supervised learning, unsupervised learning, reinforcement learning.8 In supervised learning, labelled data is fed into the learning algorithm and the model adjusts its weights unit the model fits perfectly on the other hand unsupervised learning uses unlabelled data. 9 It detects patterns or groups of data in the given information without any human input. Reinforcement learning is slightly different from supervised learning, the algorithm is not trained with labelled data rather this model learns through trial and error and based on the output the algorithm is rewarded or punished. Deep Learning is a domain of machine learning that simulates the way human brain gains knowledge. It is essentially a neural network with many layers of inter-connected nodes, which learns from large amounts of data.10 To train the neural network, large amounts of data is passed as input and the output of neural network is checked for cost/loss/error function. The weights and bias of neural network are adjusted to minimize the loss function. Back-propagation is used to update the weights using gradient descent.9,10