Artificial Intelligent Embedded Doctor (AIEDr.): A Prospect of Low Back Pain Diagnosis

Introduction

The major technique of artificial intelligence (AI) have been used in many research papers for diagnosing diseases (Das, Sanyal, & Datta, 2020) but a major problem is that they cannot be implemented in hardware. It is well known that medical diagnosis involves the use of various biosensors, this data needs to be collected from them and as such the conventional techniques of AI like support vector machines (SVM), neural networks (Das, Dey, Pal, & Roy, 2015) become inefficient with the limited memory of the microcontrollers. The physical backbone of a human is modeled with the concept of physics. In the work, an algorithm is proposed that works perfectly in hardware and the main essence of the process is that training is done separately in software and parameters are imported in the hardware to make an embedded doctor that works at any place and any time. The literature survey done in the corresponding section explores the key factor that is important for analysis is the relationship of galvanic skin response (GSR) to pain and correspondingly the method adopted by previous authors in extracting a particular feature from GSR. Then the methodology section is based on a firm mathematical model developed and from that model, the algorithm used for a diagnosis takes the shape, following which the cost-effective method of collecting GSR data is presented along with the schematic diagram and calculations. Finally, the results section is split into two parts one describing the data collected and proposing some hypotheses along with its verification. The second part describes the results of the algorithm proposed. Finally, a conclusion is made regarding the future prospects of this work. The proposed method learns by example from provided data using nonlinear regression. The most interesting fact is that if a low pass filter is used, some of the information that may be present may be lost. Therefore, the algorithm proposed herein works with the unfiltered data but still yields very accurate results and is able to differentiate noise.