A couple months ago, I wrote a blog post titled “Expanding the Scope to Neurobiology” where I talked about the really exciting work I got the opportunity to be a part of with a grad student at UC Berkeley. At the end of that post I mentioned a little bit about my science fair project I am working on this year. I wanted to take this blog post to detail what it’s all about and my progress so far! Most likely, around March or April, I will make a new blog post detailing the final product and my next steps. 

My science fair project is called “Autisight” (could change by the time I submit it), and the goal is to create a ML model that can help professionals diagnose ASD (Autism Spectrum Disorder) using MRI brain scans. So far, I have been able to successfully process the data and I am starting to run some very basic ML models such as Logistic Regression, Support Vector Machines, and Random Forests, on the data that we have. Obviously, these being very simplistic models, they produce an accuracy of about 56%, which is not even close to what we need it to be. Over the course of the next few days, I am going to start using CNNs (Convolutional Neural Networks) and adjust the hyperparameters for each neural network layer until I get a significantly higher result. After successfully creating a CNN model that works with these MRI scans, I will record accuracy metrics such as precision, recall, and loss. 

I hope this project can in the future assist medical professionals in making diagnoses. As I have been doing literature review, I have come across statistics that show how underdiagnosed ASD is, especially in young girls (source). One of the main reasons for this is because we have preconceived notions of what autism should look like, and these harmful notions were based off patterns seen in young males. Thus, when females who do in fact have ASD exhibit different behaviors or don’t exhibit what we expect, we either misdiagnose or flat-out don’t diagnose them, which further perpetuates the ASD stereotypes that exist in our society. ASD is often misdiagnosed as bipolar disorder or depression, especially in women, leading to wrong treatments that eventually harm the patient in the long run. I hope that this model can not only help professionals in diagnosing, but also aid researchers who are studying developmental disorders. By using heatmaps to unlock the “ Black Box” that CNNs tend to be (where we don’t really know why it classifies the way it does, we just know that it does), I hope to share what features the model is picking up on and consult professionals as to what they think these features represent in the real human body. 

That’s all I have for this blog post! Thank you so much and I hope to see you again soon!