Hariri Sponsored Student Led Workshop

Boston University Hosts Student-Led Workshop on Machine Learning and Brain-Behavior Data Sponsored By Hariri Institute Focused Research Program

Unlocking Novel Data Science and AI Approaches for Brain Health and Disease

On October 2nd 2023, members of the BU Center for Brain Recovery and Cognitive Neuroimaging Laboratory conducted a trainee-led tutorial to unite students and postdocs with interests in the application of machine learning techniques.

Speakers of the workshop featured current postdocs and graduate students from Boston University specializing in machine learning and fields related to brain health, behavior data, aphasia, and neuroscience. The panel featured Alan (Hantian) Liu (MS Computer Science), Manuel Marte (MS, CCC-SLP), Claire Cordella (PhD, CCC-SLP), all of whom are affiliated with the BU Center for Brain Recovery, as well as Caroline Ahn from the Graduate Program for Neuroscience at BU.

Alan Liu’s presentation laid the foundation for understanding Machine Learning Fundamentals. He delved into key concepts, including supervised and unsupervised learning, training, validation, and test sets, as well as the perils of underfitting and overfitting. Alan also discussed various evaluation metrics for models, encompassing regression, classification, IoU score, and silhouette coefficient, among others. Alan leveraged this knowledge to introduce the real-world applications of machine learning in Case Study 1, as he discussed liver fibrosis detection using a convolutional neural network. He explained that machine learning played a pivotal role in image interpretation and diagnostic assistance. Convolutional neural networks, in particular, were highlighted for their ability to extract abstract features from images through convolution, enabling image classification and segmentation.

Manuel Marte explored Case Study 2, titled “Predicting Treatment Response in Bilingual Aphasia: An Exploratory Machine Learning Approach.” He highlighted the necessity and advantages of using machine learning in scenarios where an aging bilingual population might be more affected by aphasia than initially presumed. Manuel detailed a machine learning method employing feature sets to build predictive models for treatment analysis. Two outcome analyses were central to his approach: identifying robust responders with a 50% change in naming accuracy from pre to post-therapy and distinguishing traditional responders from non-responders with a 25% change in naming accuracy. By leveraging traditional statistical models, ensemble techniques, advanced classification methods, and neural networks, Manuel harnessed machine learning to assess recovery outcomes, extract valuable demographic insights, and examine the distinctions between aphasia in bilingual and non-bilingual individuals.

Caroline Ahn, representing the Cognitive Neuroimaging Lab at Boston University, presented Case Study 3, titled “Studying the Gap Between Human and AI Abstract Reasoning.” She delved into the pursuit of developing more intelligent Artificial Intelligence programs by gaining insights into how humans construct their reasoning and decision-making processes. This research is pivotal in addressing the current gap in AI’s ability to produce results in novel situations, think intuitively, act abstractly, and exhibit emotional intelligence.

The final presenter, Claire Cordella, discussed Case Study 4, titled “Connected Speech Analysis in Post-Stroke Aphasia.” Claire emphasized the pressing need for accessible recovery programs since many individuals lack access to Speech-Language Pathologists who can provide comprehensive diagnostic assessments, meaningful subgroup diagnoses, and personalized treatment recommendations. Moreover, there’s a demand for objective, scalable methods that can be employed for efficient diagnoses, which clinicians can interpret to craft recovery plans for post-stroke aphasia patients. Claire’s research represented a substantial step in the right direction, as it demonstrated how objective connected speech measures can reasonably approximate gold-standard clinician diagnoses. This implies that the concept of AI-based diagnostic tools for post-stroke aphasia recovery plan development is entirely feasible.

This student-led seminar was a huge success, participants were able to listen to the four case studies over the course of an hour, with a second hour dedicated to questions and further tutorials when requested. Notably, Alan Liu provided an additional coding demonstration to the group towards the end of the seminar. Collaborative workshops like these foster rapid advancements in the field of machine learning and brain recovery, as students can share ideas and work together in an inspiring environment led by their peers. This outstanding event was made possible through sponsorship by Boston University’s Rafik B. Hariri Institute for Computing and Computational Science & Engineering, Digital Health Initiative’s Focused Research Program.

For more information about the Hariri Institute, visit their website.

Want to watch the workshop? Check it out on our youtube channel!