The main focus of Artificial Intelligence (AI) is to improve human cognitive abilities. It brings tremendous changes in health-care, and the data that are constantly being generated and the advances in the techniques to analyze them are the factors governing the backbone of development in AI. Today, AI is playing a vital role in the advancement of the field of medicinal diagnosis. A review of AI applications in health care and medical diagnosis with its future applications and implications The techniques are being applied to structured as well as unstructured medical data. Popular and effective AI systems include machine learning techniques such as neural networks, classical support vector machines and deep learning for structured data and NLP for unstructured data. The major Artificial Intelligence (AI) techniques include medical diagnosis including cancer, neurology and cardiology.
Introduction
There is a strong possibility that Artificial Intelligence (AI) will help doctors make more reliable and error-free clinical decisions and in some cases make it more important than human judgment in some critical areas of medical care (such as oncology). The four most important perspectives of medical investigators: a) Need for AI application in healthcare b) Data types to be analyzed by Artificial Intelligence (AI) systems c) Generating meaningful results using Artificial Intelligence through procedural mechanisms d) Disease groups that can be dealt with by AI systems.
Need for Artificial Intelligence
Artificial Intelligence (AI) in healthcare employs procedural algorithms to process large sets of datasets, and it uses the results to reach a final decision to aid medical diagnosis. The ability to learn and self-correct can be added to improve its accuracy. Moreover, Artificial Intelligence (AI) systems can provide updated medical information from journals, clinical practices with relatively less error.
Datasets
In the diagnosis phase of clinical datasets, a large portion of the data for literature analysis of artificial intelligence is obtained from electro-diagnosis, diagnosis imaging, and genetic testing. An example is, Topol and Jha encouraged radiologists to adopt artificial intelligence techniques when analyzing the huge data obtained from diagnostic images. Lee et al attempted to analyze long non-coding RNAs and the abnormal genetic expression derived from them to diagnose gastric cancer. An electrodiagnosis support system was developed by Shin et al which is used to localize injury on the brain.
There are two other major data sources which include the results obtained from laboratory tests as well as notes made from physical examination. This can be identified through images, electrophysiological, and genetic datasets as they contain a large amount of clinical notes that cannot be analyzed directly. Hence the main focus of the Artificial Intelligence (AI) system is to convert these unstructured texts into system-understandable electronic medical data.
AI Systems
Artificial Intelligence (AI) systems are divided into two major categories based on the dataset. The first category uses machine learning techniques to analyze structured data directly. The second category on the other hand works on unstructured data using natural language processing methods to enhance the structured data. For example, in the medical application of Artificial Intelligence (AI) , ML techniques work on the symptoms of patients to cluster them or to predict the probability of a particular disease, while NLP attempts to convert unstructured data into structured data which is further studied using machine learning techniques.
Disease Focus
Although there has been a rise in the literature on AI in healthcare, the main focus of research is limited to a few important disease types such as cancer, cardiovascular diseases, and neurological diseases. The main focus of AI in healthcare remains on a few diseases such as cancer, neurological diseases, and cardiovascular diseases. These diseases contribute highly to global mortality and require immediate diagnosis and treatment. It employs better analysis processes. Apart from the three main diseases, Artificial Intelligence (AI) has been applied to other diseases as well. Two examples of recent applications are the analysis done on eye image data by Long et al for diagnosis of congenital cataract disease and the other is the detection of diabetic retinopathy through retina fundus photographs by Gulshan et al.
Current Applications of AI in Medical Diagnostics
A large portion of the current machine learning analytical applications seem to fall under these classifications:
1. Chatbots
Companies are using Artificial Intelligence (AI) -chatbots with speech recognition facility that can recognize patterns through the symptoms told by the patient and draw a possible conclusion, so that the disease can be avoided and also suggest appropriate action.
2. Oncology
Scientists are trying to identify cancerous tissues using deep learning technology at a level comparable to any trained doctor, which can rapidly diagnose cancer at an early stage itself.
3. Pathology
Pathology is an important field that can be greatly helped if covered by Artificial Intelligence (AI) systems. It is the science or study of the origin, nature and course of diseases which is analyzed through laboratory tests of bodily fluids such as sputum, blood and urine and through tissue analysis. Traditional methods of diagnosis involve the use of microscopes which can be time consuming and sometimes prone to error. AI techniques such as machine learning and machine vision techniques enhance the traditional methods used by pathologists.
Rare Diseases
There are certain genetic diseases that show phenotypic aberrations. This rare disease can also be tackled by Artificial Intelligence (AI) systems that involve a combination of facial recognition software along with There are certain genetic diseases that show phenotypic aberrations. This rare disease can also be tackled by Artificial Intelligence (AI) systems that involve a combination of facial recognition software along with deep learning technology that analyzes the photos of patients and aids in detecting such rare genetic diseases.
Future of AI in Medical Diagnosis
The trend of doctors is moving away from the traditional method towards the use of AI chatbots, which is the AI form of the doctor and a UK-based digital healthcare organization is studying this collaboration of patients with AI doctors. AI is becoming popular in medical services and is in the process of evolving the traditional method to diagnose and treat diseases in a new way through the fast-growing field of machine learning and big data analytics. As Dr. Joseph Reger, CTO of Fujitsu EMEIA, said, “The news that Babylon Health has raised close to £50M to create an ‘AI doctor’ is a promising development for the healthcare industry; trials are currently underway in London, where Babylon’s technology is being used as an alternative to the non-emergency 111 number.” China has the highest rate of lung disease. InferVision’s AI innovation is being used by radiologists at Shanghai Changzheng Hospital in China.
This marks an increase in medical analysis to identify suspicious lesions and lumps in lung cancer patients. The key to better outcomes in treating a disease is its early diagnosis. The organization has integrated a computerized tomography (CT) with Artificial Intelligence (AI) that considers various properties of lung cancer and subsequently distinguishes the predicted characteristics of cancer through various CT scan data. InferVision founder and CEO Chen Kuan said in a statement that this technology will not replace doctors in any way. “It aims to eliminate excessive repetitive work and empower doctors to help them deliver faster and more accurate reports,” Reger said.
“The process of machine learning is supposed to save time, but it will only succeed if data is implemented as the lifeblood of the system,” says Fujitsu’s Reger. “In this instance, data will enable Artificial Intelligence (AI) machines to learn and understand new medical tasks, and then critically provide humans. Doctors have the information they need to diagnose problems,” Reger said. “The potential application of AI in health care can extend to predicting future diseases before they even appear, thereby improving the quality of services for patients. All of this will not be achieved without a large amount of data, the acceptance that AI will complement jobs, not replace them, and without overall investment in technology.”
AI Applications in Stroke
Stroke is a very common and frequently encountered disease, affecting more than 500 million people worldwide. It is the leading cause of deaths in Central Asia and one of the leading causes in North America. As a result, studies to prevent or treat stroke are of great help. Recently, AI has made progress and studies show that AI technology helps in three primary areas of stroke care, namely • Predicting and diagnosing stroke in advance • If diagnosed, its treatment • Outcome prediction and prognosis evaluation.
Early diagnosis
When the blood flow towards the brain is impaired, it results in cell death, which ultimately causes stroke, which is usually caused by the formation of thrombus in the vessels, hence called cerebral infarction. Many patients do not receive timely treatment due to late detection of stroke symptoms. The tool was developed by Villar et al which detects movement and predicts the occurrence of stroke. It works on the principle of two important ML algorithms i.e. PCA and Genetic Fuzzy Finite State Machine. The detection process is based on motion pattern recognition. Whenever there is a deviation from normal in the motion pattern, it is marked as an alert for stroke and inspected as soon as possible. Neuroimaging methods like CT scan and MRI are essential for disease evaluation in case of stroke diagnosis.
Some researchers have employed ML techniques to analyze neuroimaging data. ML techniques when applied to CT scans help in detecting free-floating intraluminal thrombus from carotid plaques. Rehme et al employed a well-known machine learning application called Support Vector Machine commonly referred to as SVM to track stroke probabilities. The accuracy rate of SVM in this case proved to be around 87%. Griffes et al used Bayes classification in the stroke identification process. The computer-aided method provided very comparable results compared to human experts dealing with it. Kamanitsas et al tried 3D CNN for lesion segmentation in multimodal brain MRI. Rondina et al performed Gaussian method regression on some sample MRI data to find a solid result of predictive features.
Treatment
As an acute action of measuring distress, visibility and survival rate are associated with the outcome of intravenous thrombolysis. A research was conducted to predict whether symptomatic intracranial hemorrhage developed in patients with tPA treatment by CT scan. The inputs used for SVM were whole brain images and the results have shown better performance than traditional radiology based methods.
Outcome Prediction and Prognosis Assessment
The prognosis of stroke and disease mortality are affected by many factors. ML methods have shown better prediction performance compared to traditional methods. To boost the clinical decision-making process, using logistic regression and analyzing physiological parameters for a time period of 48 hours after stroke occurrence, a model was proposed by Zhang et al. to predict the 3-month outcome after treatment. A database was compiled by Asadi et al. which included 107 patients who underwent intra-arterial therapy for acute posterior or anterior circulation stroke. The authors have studied the data obtained by SVM and simulated neural networks, and achieved an accuracy of more than 70%. They also used ML strategies to identify factors influencing the outcome of brain aneurysm for which endovascular embolization can be used for treatment. The standard regression detection model could only accomplish a 43% precision ratio but their techniques worked excellently with a precision of 97.5%.
Conclusion And Discussion
An effective Artificial Intelligence (AI) framework should have an ML component to handle structured datasets (genetic information, EP information, images) and an NLP section to extract unstructured data. At that point advanced algorithms should be created using healthcare data so that it can help doctors analyze the disease and propose treatment methods. Despite the fact that Artificial Intelligence (AI) innovations are getting liberal consideration in medical research, implementation in real life still faces hurdles. The first problem arises from regulations. There is a need to survey the safety and maintain the adequacy of AI frameworks, which is lacking in the current regulations. The US FDA has made a primary effort to give rules to evaluate AI systems to overcome the problems.
The second problem is information exchange. To work properly, Artificial Intelligence (AI) frameworks must be (continuously) fueled by information from clinical examinations. After the initial rain when the AI framework is deployed, information supply becomes a significant concern for constant advancements and changes in the framework. The incentive to share framework-related information is not given by the current healthcare situation. By the way, a healthcare revolution is yet to come in the United States to empower information being shared. Although there is a lot of potential, AI in medical diagnosis is still a modest new approach, with many physicians still yet to be convinced about its unwavering quality, efficacy, and how it will be seamlessly integrated into clinical practice without diminishing clinical expertise.
Read Also:
- Challenges And Future Of Adoption Of Artificial Intelligence (AI) In Educational Sectors
- Potential Of Artificial Intelligence (AI) In Healthcare
- From Admission To Discharge, How Artificial Intelligence (AI) Can Optimize Patient Care
- Advantages And Disadvantages Of The Use Of Artificial Intelligence (AI) In Management
- Artificial Intelligence (AI) Applications In Medicine
Leave a Reply