BCH AI and Machine Learning Working Group

Contemporary Symbolic Regression Methods for Interpretable Machine Learning

Speaker: William La Cava, PhD, Faculty, Computational Health Informatics Program at Boston Children's Hospital

Date: September 17, 2021 at 09:30AM - 10:30AM

Most interpretable machine learning research focuses on explaining the outputs of black-box models. A different, and promising, approach is to use machine learning to find the simplest possible model that meets certain performance criteria; this is the pursuit of symbolic regression. In this talk I will discuss the concepts of interpretability and explainability, and how they are used in the machine learning world. I will then discuss a pre-print that will be published in the Neurips Datasets and Benchmarks track later this year. In it, we attempt to benchmark many different approaches to symbolic regression on hundreds of problems in order to determine the strengths and weaknesses of current methods. I will discuss what lies ahead and implications for how clinicians and patients receive and process models that increasingly appear in the health system.   This event is only open to Boston Children's staff. If you would like to attend the Zoom details, please email CHIP@childrens.harvard.edu. 

Valid Inference After Hierarchical Clustering

Speaker: Lucy Gao, PhD, Assistant Professor of Statistics at the University of Waterloo

Date: May 11, 2021 at 2:00PM - 3:00PM

Dr. Gao will discuss the following article: Gao, Bien, and Witten (2020). Selective inference for hierarchical clustering. arXic:2012.02936. This journal club is only available to the BCH community. If you would like to be sent a calendar invite please email chip@childrens.harvard.edu. 

BCH AI and Machine Learning Journal Club: Andrew Beam, PhD

Speaker: Andrew Beam, PhD, Assistant Professor, Department of Epidemiology at the Harvard T.H. Chan School of Public Health

Date: April 13, 2021 at 2:00PM - 3:00PM

Dr. Beam led a discussion on the following article: Tom B Brown, Benjamin Mann, Nick Ryder, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165 [cs], 2020. Dr. Beam also discussed results from his group that evaluates this model on medical applications. 

Real-world COVID-19 Vaccine Effectiveness and the Mass Vaccination Experience in Israel

Speaker: Ben Reis, PhD, Director, Predictive Medicine Group, Computational Health Informatics Program (CHIP), Faculty at at Harvard Medical School

Date: March 16, 2021 at 2:00PM - 3:00PM

Dr. Ben Reis will lead a discussion on the recent New England Journal of Medicine paper he co-authored, providing the first real-world study of effectiveness of the Pfizer-BioNTech COVID-19 vaccine. It was the largest study yet to quantify the impact of the vaccine outside the confines of a clinical trial. The study used innovative epidemiological methods to analyze vaccine effectiveness for preventing symptomatic diseases, severe illness and death. Dr. Reis will discuss his study and the lessons learned from the nation-wide mass vaccination experience in Israel. The study has been featured in The New York Times, Bloomberg, and Fortune.

BCH AI and Machine Learning Working Group Journal Club: James Diao

Speaker: James Diao, MD, Harvard Medical School MD Student at Boston Children's Hospital

Date: February 23, 2021 at 2:00PM - 3:00PM

James will lead a discussion on approaches to addressing racial equity concerns with clinical algorithms, including for arthritis severity (Pierson et al. 2021) and kidney function estimates (Diao et al. 2021):

Let us get practical: developing and disseminating AI research and workflows to audiences of researchers and clinicians within BCH using the ChRIS platform

Speaker: Rudolph Pienaar, PhD, Staff Scientist at Boston Children's Hospital

Date: January 29, 2021 at 09:30AM - 10:30AM

We are often wowed by the *potential* of AI (and frankly other sophisticated computational approaches) to transform research and clinical workflows. New approaches seem to magically hold unbounded promise. Yet, there is often a large gulf between theory and practice, between a shiny new technique and having anyone just use it. The questions of "How do I get this ? How do I get my data from PACS to connect to this? How do I go from DICOM to something that the neural network wants? How do I get results that are useful?" In this talk I will provide some insights into practically developing, using, and disseminating "AI" (and other) workflows in the BCH clinical and research environment.

BCH AI and Machine Learning Journal Club: Guergana Savova, PhD

Speaker: Guergana Savova, PhD, Associate Professor of Pediatrics, Computational Health Informatics Program at Boston Children's Hospital

Date: December 8, 2020 at 4:45PM - 5:30PM

Dr. Savova led a discussion of tasks and applications of clinical Natural Language Processing (NLP) in medicine, such as: The landscape of neural approaches and clinical NLP (Wu et al, 2019; https://pubmed.ncbi.nlm.nih.gov/31794016/) Data challenges in clinical NLP (de-identified data, usability and challenges) Some tasks and applications Information extraction for cancer surveillance (DeepPhe-CR) (Savova et al, 2017; https://pubmed.ncbi.nlm.nih.gov/29092954/) Treatment information extraction (Bitterman et al, 2020 https://www.aclweb.org/anthology/2020.clinicalnlp-1.21.pdf; Lin et al, 2020 https://www.aclweb.org/anthology/2020.louhi-1.12.pdf) What is trending.

BCH AI and Machine Learning Journal Club: Danielle Rasooly, PhD

Speaker: Danielle Rasooly, PhD, Postdoctoral Fellow, Computational Health Informatics Program at Boston Children's Hospital

Date: November 10, 2020 at 4:45PM - 5:30PM

Dr. Rasooly led a discussion of the following paper about Google/DeepMind's AI system for breast cancer screening: McKinney et al. International evaluation of an AI system for breast cancer screening. Nature2020. as well as the following paper AI transparency/reproducibility: Haibe-Kains et al. Transparency and reproducibility in artificial intelligence. Nature 2020. ​The two papers are accessible as pdfs here.

The Age of Predictive Medicine

Speaker: Ben Reis, PhD, Faculty, Computational Health Informatics Program (CHIP); Director, Predictive Medicine Group, Computational Health Informatics Program (CHIP) Assistant Professor of Pediatrics, Harvard Medical School at Boston Children's Hospital

Date: October 16, 2020 at 09:30AM - 10:30AM

Dr. Ben Reis discussed recent developments in machine learning approaches to some of the grandest challenges of human health, including pandemic prediction, suicide prevention, bioterrorism detection, and drug safety prediction. The focus was on understanding both the methodological challenges involved and the ramifications of generating actionable predictions in these critical areas. The talk concluded by formulating a set of central challenges and opportunities facing the field of Predictive Medicine.

BCH AI and Machine Learning Working Group Lightning Talks

Date: September 9, 2020 at 09:30AM - 10:30AM

The BCH AI and Machine Learning Working Group held our first Lightning Talks session, where multiple investigators gave brief overviews of numerous Machine Learning applications at Boston Children’s Hospital to foster clinical and machine learning collaborations across the hospital.

A Gold Mine of Potential: Predictive Analytics Using Boston Children’s Hospital’s “Children’s 360” Data Warehouse

Speaker: Jonathan Bickel, MD, MS; Ronald Wilkinson, MA, MS, CBIP; Ashley Doherty, MS, at

Date: August 14, 2020 at 09:30AM - 10:30AM

Boston Children’s Hospital data warehouse integrates 15 years of extensive clinical and administrative data sources and more years of selected data sources. While the contents are used extensively for daily operational reporting, the potential for extensive retrospective and predictive analytics is largely untapped. Jonathan Bickel, Ashley Doherty, and Ron Wilkinson will show something of the breadth of data available in the EDW, discuss how predictive modeling tools can access the data, discuss ideas for predictive modeling applications that they think would be valuable, and explain the conditions on which access to the data can be granted.

AI in 3D Medical Images: Concepts, Milestones, and Opportunities

Speaker: Yangming Ou, PhD, Assistant Professor of Radiology; Affiliate Faculty, Computational Health Informatics Program; Faculty, Fetal-Neonatal Neuroimaging Data Science Center at Boston Children's Hospital

Date: July 17, 2020 at 09:30AM - 10:30AM

Dr. Yangming Ou briefly reviewed some major concepts and milestones of AI in medical images. The focus of Dr. Ou’s talk was on 3D medical images, for AI’s application in disease diagnosis, outcome prediction, early screening, neuroscience, and others. Dr. Ou then discussed some major challenges and potential opportunities, including further improving accuracy in detecting small diffuse lesions, and facilitating AI in small sample sizes.

BCH AI and Machine Learning Journal Club: Tim Miller, PhD

Speaker: Tim Miller, PhD, Assistant Professor of Pediatrics, Computational Health Informatics Program at Boston Children's Hospital

Date: June 30, 2020 at 4:45PM - 5:30PM

Dr. Timothy Miller discussed articles that he recently published on natural language processing of computerized text. 1. Dligach D, Majid A, Miller T. Toward a Clinical Text Encoder: Pretraining for Clinical Natural Language Processing With Applications to Substance Misuse. SSRN. 2020. 2. Miller T, Avillach P, Mandl K. Experiences Implementing Scalable, Containerized, Cloud-based NLP for Extracting Biobank Participant Phenotypes at Scale. SSRN. 2020.

BCH AI and Machine Learning Journal Club: Arjun Manrai, PhD

Speaker: Arjun (Raj) Manrai, PhD, Faculty, Computational Health Informatics Program (CHIP); Director, Laboratory for Probabilistic Medical Reasoning; Assistant Professor, Harvard Medical School at Boston Children's Hospital

Date: May 8, 2020 at 09:30AM - 10:30AM

Blood laboratory measures such as glucose and hemoglobin are the basis for much of clinical decision making, yet baseline variation for many laboratory measures remains incompletely characterized across age, gender, and race groups. I will introduce foundational techniques from machine learning and statistical genetics and show how they can be applied to systematically unpack variation in blood laboratory data across population groups. These analyses reveal widespread demographic structure in blood laboratory data.