University:
Boston University
Course:
SAR HP 151 | Introduction to Health Professions
Academic year:

2022
Views:

275

Pages:

15
Author:

gremijimaipmw

Artificial Intelligence for Biomedicine and Healthcare Agenda  Welcome Remarks  Presentations Boston University Slideshow Title Goes Here  Yannis Paschalidis  Margrit Betke  Mark Kramer  Kayhan Batmanghelich  Vijaya B. Kolachalama  Adriana Tomic  Daniel Segrè  Ji-Xin Cheng  Emma Lejeune  Lei Tian  Closing Remarks COVID and Preventing the next Pandemic Predictive Models for Level-of-Care Requirements • Worked with datasets from Wuhan, Mexico, Brazil, MGH-Brigham, BMC (n=900−113,000) • Predicting Hospitalizations, ICU Admissions, Need for Ventilation, Mortality • BMC cohort revealed role of Social Determinants of Health and Racial Biases Boston University Slideshow Title Goes Here A multidisciplinary team from Hariri, CISE, CEID, NEIDL, and BTEC, working on an NSF-funded project to develop models that predict disease emergence and spread, and to devise pandemic mitigation strategies Alzheimer’s Disease & Dementia Boston University Slideshow •TitleMost Goes Here common form of dementia, > 5M in the US with AD ($300B) → 14M by 2050 ($1.1T) • Most clinical trials fail Digital Biomarkers Automated Cognitive Assessment Pipeline • Leveraging Automated Speech Recognition (ASR) and Natural Language Processing (NLP) • Does (almost) as well as clinicians (93% AUC) Predictive and Prescriptive Modeling Boston University Slideshow Title Goes Here Key Challenges in Health Applications • Interpretability & Robustness • Statistical ML & Optimization Predictive Models of Pregnancy • SPH PRESTO (Pregnancy Study Online) dataset • eIVF dataset ESHRE 24hr Tweetutorial Prescriptive Models for Hypertension • Launching a BMC Pilot Study • AI model to make personalized recommendations for Hypertension drugs AI for Pediatric US Diagnosis, AI for Physical Therapy, AI for Post-Stroke Prognosis AI to Interpret Ultrasound Videos for the Diagnosis of Pediatric Pneumonia in Limited Resource Settings (here Lusaka, Zambia) Pneumonia is leading cause of death in children < 5 years worldwide with mortality 68% at 6 month Goal: Develop AI cell phone app for first-line health care workers to use mobile US device to image infants & diagnose and treat pneumonia. Results: Collected 12 videos/infant of 2 cohorts of 200 infants. Developed AI models for single image analysis. Low F1 score. Ongoing Work: Developing AI model for US video analysis. ExerciseCheck Home-based Physical Therapy with AI Results: App for managing logistics. Initial experiments. Ongoing Work: Developing attention-based deep-learning model for 3D human pose tracking. AI to Predict Response to Post-Stroke Rehabilitation Stroke is a leading cause of severe and complex long-term disability. Aphasia affects approximately 1/3 of stroke survivors. A personalized prognosis on the evolution of aphasia provides guidance for clinicians to select appropriate treatment. Results: Developed support vector machine and random forest models to predict responsiveness to treatment using pretreatment behavioral, demographic, and structural and functional neuroimaging data. Best model’s prediction score F1=0.94. Ultimate Goal: Recommend treatment Ongoing Work: Developing traditionally-learned & deep-learned models for patient outcome prediction. AI to Predict Response to Post-Stroke Rehabilitation Ongoing Work: Develop fusion model. Examples of Machine Learning in Biomedicine: Applications to Epilepsy Epilepsy impacts brain dynamics in subtle ways ... spike ripple spindle X Bad X 0.05 s attention ? 10 s X memory X IQ Detect these subtle dynamics CNN X So now what? after detection, do something ... X attention memory IQ Close the loop Practical problems (in human health) - motivate mathematics & statistics - to impact these problems.