Leo H​. '24, driven by his interest in artificial intelligence (AI) and computer vision, has​ recently gained recognition for his innovative work in dementia prediction. Collaborating with a doctoral student from Cambridge University, Leo incorporated metadata into his approach, achieving significant results.

Leo connected with the doctoral student through the Horizon Inspires Program. Horizon Academic (HARP), as described on their website, is a trimester-long online research program for extraordinary high school students aimed at refining their interest in an academic subject. Within this program, students have the opportunity to develop a college-level research project under the individualized guidance of a professor from a globally renowned university. 

Their joint project focused on predicting dementia using MRI brain scans as input data, a vital step in early detection and intervention for a condition with no known cure. What sets Leo's research apart is his application of "self-supervised" learning, a novel training method in the field of AI. Unlike conventional approaches that rely heavily on labeled data, self-supervised neural networks do not require explicit labels for input data. Instead, they identify patterns, differences, and similarities within the data. Leo's innovation was to use "weak labels" derived from metadata associated with MRI scans, such as age, sex, dominant hand, and mini dementia test scores, to train the model. This approach allows the AI to learn from a wealth of information and bridges the gap between AI and classical diagnosis methods.

The integration of metadata into his research was not without its challenges. Traditionally, medical datasets have struggled to effectively utilize metadata during the training process. However, Leo's method has shown promise in harnessing this valuable information to enhance prediction accuracy.

When asked about his future plans, Leo expressed a strong commitment to continuing his exploration of AI and computer vision. He intends to pursue undergraduate research in AI during college and further expand his understanding of the field by taking relevant courses. With a clear vision of improvement opportunities for his research, including applying it to larger datasets and other data-rich medical fields, Leo's dedication to advancing AI-driven healthcare solutions is apparent.

Reflecting on his experience, Leo shared, "In this project, I gained a much deeper understanding of AI and computer vision. Now I am truly drawn to the beauty of AI."

 

Read Leo's paper here