Deep Learning for Medical Care

Abstract

Deep Learning is a computer model trained to achieve classification from the  input data like  images, text and audio.Machine Learning uses algorithms to learn from data to cognizant the decision from what it is learned.Whereas, the DL model creates Artificial Neural Network layer by structured algorithms to make smart decisions.Convolutional Neural Network or ConVet is a class of DNN (Deep Neural Network) energized by biological processes by implementing the connectivity pattern between neurons featured the organization of the animal visual cortex. Deep learning in medical care is a fast , effective and accurate mathematical model designed to function as a human brain.The accuracy of deep learning is highly appreciable in wide applications as they process aggrandize data and the principle of learning from previous output to distill their competency to make correlations and connections.The computing capability of deep learning models with multiple layers solves critical problems and proven to be promising diagnosis in the field of healthcare by researchers and healthcare professionals.

Introduction

Deep Learning , a deep structured learning is a subset of Artificial Intelligence and type of Machine Learning used to analyze the data.Deep learning is becoming popular with its successful experimental result and also has potential to change the future of medical care in diagnosis and treatment of deadly disease.The model is capable of performing speedy data analysis with high accuracy.The recent survey from Report-linker shows that the AI medical care market will grow to $36 billion by 2025 from $2.1 billion in 2018.

        Figure 1:Architecture of Deep Learning                                       Figure 2: Deep Learning type of Machine Learning and subset of AI

Many AI startups have emerged in the field of medicine for its automation and diagnosing accuracy. The start-ups that support Medical Intuitions are Ada, Lunit, Sense.ly, Insilico Medicine, PathAI, Aira and so on. The deep learning model filter the data’s that passes through each consecutive layer in multiple hidden layers by using the result from the past one to inform its output. The Deep Learning is generally inspired by the function of biological neurons which connects with one another to process information in the brains of animals .Similarly, in Deep Learning, Artificial Neural Networks is used to perform a biological neurons function in the system. Hence the hidden layer in deep learning will execute the mathematical translation in-order to make raw input data into meaningful output data.

Deep Learning in Medical Care

Electronic Health Record

The need of the HITECH system in the field of medical care is highly essential to understand health data in a Biological System (BS), Electronic Health Record (EHR), Medical Image (MI) and Physiological Signal Processing (PSP). The role of Deep Learning in Electronic Health Record (EHR) is to maintain the medical history of the patient. The patient data like demographic information, medical history records, and lab results are stored in the EHR system and these data are used to identify the risk pattern of the patient to derive the conclusion. The EHR system data are used in two ways, static prediction and prediction based on a set of inputs. A static prediction is accomplished by collecting and feeding the data set from the researchers into the system and embeds the code from International Statistical Classification of Diseases and Related Health Problems (ICD). Based on patient history and hospital visits the system will predict the probability of heart failure or any cause of disease. The second way is prediction based on a set of input from the EHR system and deals with individual input or entire input using an advanced DL model with ANN (Artificial Neural Network) .The method will predict when will be the next visit to hospital and incite reason for the visit from ICD codes information amassed from a patient’s latest visit to hospital and the time glide because of the patient’s recent visit. The EHR system data is bantam difficult when the physician tries to diagnose infrequent diseases with limited data set.

Generative Adversarial Network

The Generative Adversarial Network (GAN), a DL model, solves the issue of EHR systems. The principle of GAN is to collect patient records, develop more datasets and train the model with available datasets. Generative Adversarial Network competes with ANN against two factors namely generator and discriminator. The generator trained to learn the particular characteristics from given data set and create new instances in-order to make the discriminator to think the generated data is genuine. After new instances created by the generator, discriminator checks both data sets for genuineness and determines the generated data is real or fake. The process is repeated till the generator is trained completely and produces genuine data sets to make the model work with quality data.

Figure:3 Performance of Deep Learning

Cancer Diagnosis

For cancer diagnosis, medical imaging like CT scan, MRI scan and X-ray are commonly used by oncologists. The most common scan in cancer diagnosis is CT scan, Computed Tomography (CT) scan which follows medical imaging procedures by considering the combination of computer-processed X-ray measurement obtained from various angles to output the cross-section image of a particular location of the scanned object. Even the traditional systems are proven to be effective in the diagnosis of cancer, a more number of patients suffer from cancer disease due to in-accuracy of these scan machines. Hence the class of DL model, Artificial Neural Network like CNNs (Convolutional Neural Networks) produces the promising results in detection of cancer. The implementation of ANNs can detect early stage cancer with less false diagnosis and also researchers proved that the detection of various kinds of cancer at earlier stage by the trained DL model is with high accuracy.

Applications of Deep Learning in the field of Medical Care:

Deep Learning methods use the data available in the EHR system to decrease the misdiagnose rate and to predict the procedure outcome. The Artificial Neural Network, ANN, guide the physician to analyze the information and observe different conditions from large datasets received from different fields of medical care like Medical Imaging. The observed conditions by ANN are:

1. Analyzing of blood sample

2. Identification of heart related problems

3. Detection of glucose level in diabetics patients

4. Detection of the tumour using image analysis

5. Cancer diagnosis and detection of cancer cells

6. Identification of osteoarthritis from patient MRI scans

Conclusion

Use of DL in medical care includes intensive tasks like training the Artificial Neural Network Model to analyze a huge amount of data set obtained from images and videos. Training of these models requires highly demandable hardwares to produce challenging and scalable output. The deadly diseases like cancer to be predicted at earlier stage to save the lives of human and the deep learning model is the proven technology for the diagnosis accuracy of diseases like  Diabetic Retinopathy ,Human Immunodeficiency Virus (HIV),different kinds of Cancer, Genome Research and so on. The automatic trained DL model will save time and human life.