Deep Learning has been rapidly adopted in various spheres of healthcare for translating medical data into improved human healthcare. Deep learning based healthcare applications assist physicians to make faster, cheaper and more accurate diagnosis. This talk will focus on how commonly occurring Dry Eye Disease (DED) can be diagnosed accurately and speedily using deep learning based automated approach.

DED is one of the commonly occurring chronic disease in the world today. It causes severe discomfort in eye, visual disturbance and blurred vision impacting the quality of life of patient. Certain factors such as prolonged use of electronic gadgets, old age, environmental conditions, medication, smoking habits and use of contact lens can disturb the tear film balance and can lead to evaporation of moisture from tear film which causes dry eye disease. If DED is left untreated, it can cause infection, corneal ulcer or blindness. However, diagnosis of dry eye is a difficult task because it occurs due to different factors. An ophthalmologist sometimes requires multiple tests or repetitive tests for proper diagnosis. Moreover, the major drawbacks of clinical diagnosis are: 1) Higher time in clinical diagnosis as it is done manually. This has severe impact during mass screening in Civil Hospitals and Multispecialilty Hospitals 2) Diagnosis is subjective in nature 3) Accurate severity level of DED is not identified and 4) Medication may be prescribed for incorrect period on the basis of inaccurate severity level. To overcome these drawbacks, we have developed a deep learning based automated approach to diagnose DED considering Tear Film Breakup Time (TBUT) which is a standard diagnostic procedure. This automated approach is to assist ophthalmologist and optometrist to bring objectivity in diagnosis, to increase diagnosis accuracy and to make diagnosis faster so that ophthalmologist can devote more time in counselling of patients.

The talk will include motivation, precise problem statement, proposed solution, data set generated consulting ophthalmologists and experimental results. Finally, the possible extensions of our work and the future scope of research in healthcare sector will be discussed.


Outline/Structure of the Talk

  • Dry eye disease: causes, symptoms and complications (2 mins)
  • Motivation and problem statement (3 mins)
  • Deep learning based solution (5 mins)
  • Dataset and experimental results (5 mins)
  • Possible future extensions of our work and research directions in healthcare (3 mins)
  • Q & A (2 mins)

Learning Outcome

  • After attending this session, the attendees will ...
    • understand how deep learning can be utilized to diagnose unexplored or rarely focused diseases.
    • understand how deep learning is useful to make faster, cheaper, and more accurate disease diagnosis.
    • be familiar with future research possibilities in healthcare sector.

Target Audience

Data Scientists, Machine Learning/Deep Learning Practitioners, Industry Professionals from Healthcare Sector, Researchers, Doctors, Students & Faculty Members from Engineering and Technology as well as Medical.

Prerequisites for Attendees

  • Basic understanding of neural networks.
schedule Submitted 1 year ago

Public Feedback

comment Suggest improvements to the Author
  • Vishal Gokhale
    By Vishal Gokhale  ~  8 months ago
    reply Reply

    Hello Dr Mehta, 
    Thanks for the proposal. 
    For the PC to get a clear understanding, request you to please elaborate a little more on the following points:

    1. Details on outline elements:

    • Deep learning based solution (5 mins)
       - Choice of the model (You may elaborate on journey in the actual talk - where you outline the challenges faced and the evolution) 
    • Dataset and experimental results (5 mins)
      - high level metrics (ex. overall Accuracy, Precision, Recall etc) 
    • Possible future extensions of our work and research directions in healthcare (3 mins)
      - Some details that provide a measure of the gap between practically usable state and current state. 

    2. The sample video uploaded is of 43 seconds only. 
    Request you to please upload longer video (3-5 mins) covering summary of the talk. 

    3. The slides are about hypothyroidism. Request you update the link.


    • Dr. Mayuri Mehta
      By Dr. Mayuri Mehta  ~  8 months ago
      reply Reply

      Hi Vishal,

      First of all sorry for delayed reply.

      Hope you are safe and healthy.

      Pls find clarification for the points as below.

      • Choice of the model
        • We have used GoogleNET CNN after experimenting with four pretrained networks: AlexNet, VGG 16, GoggleNet and ResNet.
      • high level metrics (ex. overall Accuracy, Precision, Recall etc) 
        • Required data (TBUT Videos) is not available publicly. So dataset is self-generated consulting different ophthalmologists. TBUT video frames are classified into one of the four classes to detect DED. Experiments are conducted in MATLAB. Results are evaluated using metrics such as accuracy, pearson correlation, sensitivity, specificity, DED detection time and severity level identified.
      • Some details that provide a measure of the gap between practically usable state and current state
        • Current clinical tests are subjective in nature, time consuming and require repetitive tests in case patient blinks eye during test or ophthalmologist is not able to identify tear film breakup area. Our proposed solution brings objectivity in the test, facilitates faster diagnosis during the mass screening of eye diseases and also identify severity level of dry eye disease. Hope my understanding of this query and reply are as per your expectation.
      • Sample Video
        • Presently we are working from home. So I won’t be able to upload video with slide projection. If it is fine without projection of slides, let me pls know. Otherwise I will be able to upload longer video along with slide projection once college resumes.
      • Slides
        • Sample slides related to DED detection have been uploaded.
      • Kuldeep Jiwani
        By Kuldeep Jiwani  ~  6 months ago
        reply Reply

        Hello Dr. Mayuri,

        Thanks for providing the details. Could you please provide some more information on the dataset. We understand that this is synthetic data but could you explain its relation with real world data. We wish to understand its impact and connect with real world use case. Let's say how ready would it be for real world usage and what are the views of ophthalmologists on using this to solving real world problems.

        • Dr. Mayuri Mehta
          By Dr. Mayuri Mehta  ~  6 months ago
          reply Reply

          Dear Kuldeep,


          Our dataset is not constructed artificially, rather constructed collecting TBUT test videos from different Ophthalmologists. It contains video of patients who went through the TBUT test. These videos are converted to frames and frames are used as training/validation and testing data. Through out this research, we are in association with some Ophthalmologists and we are sincerely thankful to them for all their help including providing data.

          Regarding real world usage and help to ophthalmologists, Dry Eye Disease (DED) is a multifactorial disease. There are many factors that can cause DED. In order to diagnose DED accurately, an ophthalmologist may require to carry out multiple tests. Among such tests, TBUT is a commonly used test by ophthalmologists. Our proposed ‘TBUT based DED detection approach’ could be deployed in a real life scenario after training it with more data. Increasing data would help to learn features of breakup more accurately and thereby to get sufficient accuracy. If used in clinical practice, it would not only reduce manual decision making time but would also increase productivity, precision and efficacy of ophthalmologists by giving them more time for patient's counselling. 

          • Kuldeep Jiwani
            By Kuldeep Jiwani  ~  6 months ago
            reply Reply

            Thanks for the information, this was very helpful

  • Dr. Santonu Goswami
    By Dr. Santonu Goswami  ~  8 months ago
    reply Reply

    Hi Dr. Mehta,

    Thank you for the interesting proposal.

    Outline of your talk and learning objectives and nicely put. You also provide an excellent background to introduce Dry Eye Disease (DED).

    But as of now, the proposal lacks clarity and detials on the deep learning aproaches that the research adopts to achieve the objectives. This needs to be updated. As of now, the slides do not provide any information on future directions of the current research.

    You mention about your talk in WiDS2020 conference. If possible, a link to the talk will provide a very good video representation as well. 

    Thank you, 




    • Raja Mohan
      By Raja Mohan  ~  8 months ago
      reply Reply

      Good work Mehta ji, I can see the work done. My request to all committee members from my observation is that, with all the respect and having myself in the journey I feel you are asking and looking for a lot in 20 min sessions here in all applicants. I can possibly imagine from the applicants end that its hard to give all details with what is happening. Also, I would stress to the committee and the voters to encourage more of application based proposals too.

      • Dr. Mayuri Mehta
        By Dr. Mayuri Mehta  ~  8 months ago
        reply Reply

        Thank you. Agree that 20 mins are too less to speak on research work (of 1.5 yrs) satisfactorily and justifiably. I shall try my best to cover each important aspect related to this  research topic within 20 mins.

    • Dr. Mayuri Mehta
      By Dr. Mayuri Mehta  ~  8 months ago
      reply Reply

      Dear Dr. Santonu,

      For the same reason that 2nd paper of this research is under process of publication and slides are publicly available, I would like to include slides related to deep learning based solution and future directions later or before conference only if proposal gets selected. However, I can surely share those slides with program committee for proposal evaluation.

      WiDS 2020 was planned during this month. However, due to COVID-19, it is postponed to May last week or June first week.

  • 20 Mins

    Healthcare industry across the world has progressed very rapidly over the last two decades. However, the healthcare industry is behind other sectors in adopting the newer IT technologies. This talk primarily focuses on imbibing Artificial Intelligence (AI) in medical domain innovations. Like other revolutionary advances in medicine, AI is to be integrated into healthcare practices.

    Healthcare using Artificial Intelligence is amongst the fastest growing research area across the globe. A massive amount of heterogeneous data generated in healthcare sector offers opportunities for big data analytics. Such analysis transforms big data into real and actionable insights to healthcare practices, thus provide new understanding and ways for better and quicker treatment and improve overall individual and population health.

    Recent advancements in AI are proving beneficial in development of applications in various spheres of healthcare such as microbiological analysis, discovery of drug, disease diagnosis, Genomics, medical imaging and bioinformatics. Due to increasing availability of electronic healthcare data (structured as well as unstructured data) and rapid progress of analytics techniques, a lot of research is being carried out in this area. Popular AI techniques include machine learning/deep learning for structured data and natural language processing for unstructured data. Guided by relevant clinical questions, automation using AI can unlock clinically relevant information hidden in the massive amount of structured/unstructured data, which in turn can assist clinical decision making.

    The talk connects three contemporary areas of research: AI, Healthcare and Bigdata Analytics. It will provide attendees a collective update on developments in healthcare using AI, major challenges, opportunities and future research directions.

  • 20 Mins

    Machine Learning has been rapidly adopted in various spheres of healthcare for translating medical data into improved human healthcare. Deep learning based healthcare applications assist physicians to make faster, cheaper and more accurate diagnosis. This talk will focus on “Early Detection of Hypothyroidism in Infants using Machine Learning”.

    Thyroid is a hormone secreting gland which influences all metabolic activities in our body. Hypothyroidism is a common disorder of thyroid that occurs when thyroid gland produces an insufficient amount of thyroid hormone. Deficiency of thyroid hormone at birth leads to hypothyroidism in babies. The common hypothyroidism symptoms in infants are prolong jaundice, protruding tongue, hoarse cry, puffy face, pain and swelling in joints, goiter and umbilical hernia. During early stage of hypothyroidism, babies may not have noticeable symptoms and hence, doctors (Physicians, Paediatricians and Paediatric Endocrinologists) face difficulty in recognizing hypothyroidism in infants. If hypothyroidism in infants isn’t treated during early stage, severe complications such as mental retardation, slower linear growth, loss of IQ, poor muscle tone, sensorinueral deafness, speech disorder and vision problem may arise. As a consequence, infant’s growth cannot be proceeded as healthy infants. To prevent such complications, we have developed a novel approach to diagnose hypothyroidism in infants during its early stage. To the best of our knowledge, this is the first attempt to detect hypothyroidism based on only facial symptoms viz. puffy face, jaundice, swelling around eyes, protruding tongue and flat bridged nose with broad fleshy tip.

    This talk will include motivation for this work, precise problem statement and its solution, data set generated consulting Pediatric Endocrinologists and experimental results. Finally, possible extensions of this work and the future scope of research in healthcare sector will be discussed.