Alladi Ramakrishnan Hall

Machine learning and predicting clinical outcomes (Pre-synopsis talk)

Chandrani Kumari

IMSc, Chennai

Predictions using machine learning techniques in the field of healthcare have gained a lot of popularity among the scientific community in the past few years. In healthcare, machine learning is being used to analyze large volumes of data, including electronic health records, medical images, and genomics data, to identify patterns, predict disease risk, and personalize treatment. This thesis focuses on using machine learning techniques to predict health outcomes.

In the first part of my talk, I will discuss a predictive model of a growing fetus. We proposed that the
Gompertz model fits fetal biometries with just three intuitive parameters and can be a basis for future growth standards. Two of these parameters - 𝑡₀ (the inflection time) and 𝑐 (the rate of decrease of growth rate) - can be treated as universal to all fetuses, while the third parameter 𝐴 can be modeled as an overall scale parameter specific to each fetus, which captures the individual variation in growth. Also, using early ultrasound data available, a regression model can be used to predict the birth weight. Finally, we show that the Gompertz growth curve is a close fit to the standards from WHO, NICHD, and INTERGROWTH.

However, there are also challenges to be addressed, including issues related to data privacy and bias in algorithm development. Data privacy is a critical concern when it comes to using machine learning in healthcare. This is because health data is often sensitive and personal, and it is important to protect patient's privacy and maintain confidentiality. In the second part of my work, I will talk about how we can generate synthetic data as this approach can be used to address these concerns.

At the end, I will discuss some other health-related machine-learning projects that I have been working on.