Artificial Intelligence in Healthcare

    Rajeshwar Reddy

    What is Artificial Intelligence (AI) and what are the problems it can solve in healthcare?

    “[The automation of] activities that we associate with human thinking, activities such as decision-making, problem solving, learning …” Machines have far superior computational abilities than humans. They can sort through enormous amounts of data and use it to make better decisions.

    AI can:

    • Find patterns, trends, and associations
    • Discover inefficiencies
    • Learn and improve
    • Execute plans
    • Predict future outcomes based on historical trends

    What are the main components of AI?

    • The core component of an AI system is an algorithm, it can be anything from finding the best liver segmentation, brain tumour segmentation to detecting the presence of Pneumonia or not.
    • A processing system in which you can actually store the algorithm and run it.
    • Training platform i.e., the algorithm has to run over and over again to get accurate results or to function properly

    Principle areas of AI application

    AI has numerous application areas where it is presenting ground-breaking human level results. All these areas are evolving on daily basis, new research innovations are occurring. The most popular among the applications is Healthcare.

    Figure 1: AI in different application areas

    Artificial intelligence is poised to make radical changes in healthcare, transforming areas such as,

    • Diagnosis
    • Genomics,
    • Surgical robotics,
    • Drug discovery

    In the coming years, artificial intelligence has the potential to lower healthcare costs, identify more effective treatments, and facilitate prevention and early detection of diseases.

    The need for AI in Healthcare

    • Specialist doctors are few in number and are overburdened.
    • One in three physicians are over 55 years of age, and a third of physicians are expected to retire in the next decade (Source: BMC Health Services Research).
    • The volumes of current patient data as well as their complexity make clinical decision making more challenging than ever.
    • Biomedical informatics methods are required to process the data and form recommendations and/or predictions.
    • AI can be a solution to explore insights from the huge amount of data and assist doctors in medical decision making.
    • Any mistake or failure in the diagnostic process leading to a misdiagnosis, or a delayed diagnosis is considered as diagnostic error.
    • An AI system can help reduce diagnostic and therapeutic errors in the human clinical practice.
    • Research shows that at least 5% of adults in the United States experience a diagnostic error each year in outpatient settings (Source: World Health Organization
    • Medical mistakes are now estimated to kill up to 440,000 people in U.S. hospitals each year.
    • AI with its ground-breaking results can help to reduce these misdiagnosis statistics.
    • AI can help in early detection of diseases, patient monitoring, predictions and resource management, and lifestyle management.
    • AI can sift through bunch of messy raw data and learn how to organize the data and predict health outcomes.
    • Research suggests that AI can be used to predict mortality, readmission, and other events that have an adverse impact on healthcare.

    What can AI do in health?

    • Diagnosing patients
    • New insights into diseases
    • Improving how hospitals run

    What are the challenges?

    • Training staff
    • Cyber security
    • Patient confidentiality

    AI in Healthcare trends 

    AIH Research and Development Focus

    1. Artificial Intelligence based Medical Image Processing

    a) Brain MRI Analysis and Tumour Segmentation using Artificial Intelligence

     Manual Segmentation Challenges

    • Time consuming
    • Tedious
    • Inter-rater variability
    • Irregular form
    • Confusing boundaries of tumour

    Figure 2: Manual Brain Tumour Segmentation (Variation in tumour size, shape, etc.)

    Automated Brain Tumor Segmentation

    • Convolutional Neural Networks achieved astonishing results for automatic brain tumour segmentation.
    • Deep CNN architectures are widely used in brain MRI for:
    1. Preprocessing MRI data
    2. Detecting and segmenting lesions/tumours
    3. Segmenting tumors
    4. Segmenting whole tissues


    Figure 3: Automated Brain Tumour Segmentation

    AI algorithms for automated brain tumour segmentation

    Significance of preprocessing of BRATS data for tumour segmentation using 3D-Unet


    b) Automated Liver Tumour Segmentation and Classification

    Liver Tumor Segmentation

    • Liver tumor segmentation is considered a more challenging task than liver segmentation.
    • Requires a pipeline of different techniques.
    • Manual methods for liver tumour segmentation are erroneous.
    • Computer-aided Diagnosis analysis is of great help to radiologists.
    • The image intensity values are truncated for all scans to the range of [-200,250] HU to remove the irrelevant details.
    • Images are resampled with the same resolution 0.69 x0.69 x1.0 mm3
    • Once the CT images are preprocessed, they are used to train AI models for automated liver and lesion segmentation



    What we need from health centers and radiologists?

    •The expert radiologists are required to manually segment the tumorus in brain MRI.
    •The expert radiologists are required to manually segment the liver and lesion in the liver CT scan.
    •The AI algorithms will be then trained on this label data to automatically segment any given tumour and perform classification.
    •Radiologist feedback and support for clinical trials of the trained AI model


    2. Development of Artificial Intelligence based Heart Sound Classification Algorithm

    Generation of Heart Sounds

    Time and Frequency Representation of Heart Sound Signal




    Methodology for algorithm development


    What we need from health centers and cardiologists?

    •We need proper-labeled datasets to train machine learning algorithms. The more proper  the  sounds are diagnosed and labelled the better/accurate algorithms can be developed
    •Cardiologists feedback and facility for clinical trails of the trained model.


    3. Artificial Intelligence based Automated Assessment of Thalassemia Minor

    Essence of the problem

    According to World Health Organization (WHO), there are at least 948,000 new carrier couples and over 1.7 million pregnancies to carrier couples annually.

    Around 75% are actually at risk. In principle, all need expert risk assessment and genetic counselling.


    For optimum treatment, every Thalassemic child will need frequent transfusions of screened packed red cells and regular iron chelation therapy.

    The treatment and Management of all Thalassemia Major Patients require a huge health budget each year, which is out of reach of our health departments and an impossible target to achieve at the present moment. This situation leaves only one alternative –Thalassemic Awareness and Prevention on national level.



    What we need from health centers and hematologists?

    •Bunch of data, i.e., electrophoresis test images are required to train AI models for automated diagnosis of thalassemia.
    •Hematologists to diagnose and report on the data.
    •Providing facility for clinical trials of the designed and trained model for electrophoresis test


    4. Some Special Research Focus and Problem Areas

    •Artificial Intelligence based Human Posture Classification and Fall Detection
    •Sound Source Separation and Hearing Aid using Artificial Intelligence models



    AI in Healthcare: Research and Development Trends

    •There is a need to incorporate AI in health services.
    •We need to learn from the experiences of the developed countries regarding challenges of the application of AI in healthcare.
    •AI experts/practitioners, doctors and health service providers need to actively and closely work together to develop innovative solutions that improve healthcare delivery.
    •In the healthcare industry, nearly 86% of the mistakes are preventable, In the next 5 years, AI health market will grow by more than 10 percent.
    •Various companies are working on ways to use AI for better blood tests. The technological solutions they create could bring about faster conclusions and require less blood from patients.