Hack Aotearoa AI in Healthcare Conference – Workshops
All workshops are available Friday and Saturday
Pitfalls of Health Data Science
Jesse Raffa (MIT), Katrina Poppe (University of Auckland), Will Scarrold (Datamine), Luke Boyle (Orion Health)
Location: Case Room 4
The increased adoption of electronic health records (EHR) has created novel opportunities for researchers, including clinicians and data scientists, to access large, enriched patient databases. With these data, researchers are in a position to approach questions with statistical power previously unheard of in medical research. In this workshop, we present and discuss challenges in the use of EHR data for research, as well as explore the unique opportunities provided by these data.
Data Mining Clinical Notes
Christina Chen (MIT), Kate Loveys (University of Auckland), Corrin Lakeland (CME Connect), Ernestynne Walsh (Nicholson Consulting)
Location: OGGB 5
There are valuable insights that can be gained from information hidden in free text clinical notes. In this workshop, participants will be introduced to the process of extracting data from free text. We will be using python in jupyter notebooks and the MIMIC-III database.
Introduction to Machine Learning in Healthcare
Jenna Wiens (MIT and University of Michigan in Ann Arbor), Yun Sing (University of Auckland), Quentin Thurier (Orion Health)
Location: OGGB 3
With the rapid advancement of technology, hospitals are creating and collecting vast amounts of data about their patients. These data, stored in the electronic health record, have the potential to impact clinical care in a variety of ways, e.g., in identifying patients at greatest risk of an adverse outcome, in predicting the efﬁcacy of a particular drug, or in matching patients with the best treatments. In this workshop, you will be working with a subset of data collected from patients admitted to an intensive care unit at the Beth Israel Deaconess Medical Center in Boston. You will develop classiﬁcation tools that automatically identify patients at risk of in-hospital mortality. Such a model could help clinicians target interventions, facilitating better management and improved patient outcomes. We emphasize that these data represent real patient admissions and real outcomes; they have been made available for research and educational purposes by PhysioNet.org. Such datasets are critical to the fundamental advancement of clinical-decision support tools.
Demystifying Healthcare Information Systems
Chen Xie (MIT), Samuel Wong & Eduardo Monzo (Vensa), Koray Atalag & Mike Merry (The Clinician), Stephen Connor (Geonostic)
Location: Case Room 2
Big data, precision medicine, cloud computing, API, data lake… What does all this mean, and what do we actually need for our overarching goal of extracting insights from healthcare data? This workshop explains the fundamental computing concepts behind medical (and non-medical) information systems, including client-server interactions, relational databases, APIs, and web services. These concepts will then be pieced together to describe how most medical applications are built and utilized.
The building process of the MIMIC database from its various sources will be explained to demonstrate how disparate data sources can be combined and utilized. Emphasis will be put on understanding concepts,and clarifying unnecessary technical jargon. As concepts are explained, attendees will demonstrate their understanding by turning raw demo data into an analyzable healthcare database, running some basic analyses, and building a web service to make the database accessible.
MIMIC-III (‘Medical Information Mart for Intensive Care’) and the eICU Collaborative Research Database
Tom Pollard (MIT), Omar Badawi & Cheryl Hiddleson (Philips)
Location: 260-040 B
MIMIC-III is a large, single-center database comprising information relating to patients admitted to critical care units at a large tertiary care hospital. Data includes vital signs, medications, laboratory measurements, observations and notes charted by care providers, fluid balance, procedure codes, diagnostic codes, imaging reports, hospital length of stay, survival data, and more. The database supports applications including academic and industrial research, quality improvement initiatives, and higher education coursework.
The second half will explore the details of the eICU Collaborative Research Database and review how a research network to advance critical care research was created across hundreds of unaffiliated hospitals. The eICU Collaborative Research Database consists of a representative sample of the eICU Research Institute data which includes granular, clinical and outcomes data from over 3.5 million ICU patients. The inclusion of detailed vital signs, archived in 5 minute median values, from bedside monitors for all patients creates a unique opportunity to evaluate vital sign trends and short-term responses to critical care treatments. Challenges in working with large, multicentre databases will be discussed and unique insights into interpreting data within the eICU Collaborative Research Database will be provided. This database is now freely available to the public and is currently used for the development of advanced clinical decision support tools as well as epidemiology and other academic research projects.
Analysing ECG using Deep Learning
Jonathon Rubin (Philips Research), Alan Williams & James Williams (Isogonal)
Location: Case Room 3
In this workshop we will develop deep learning-based classifiers to process and analyze electrocardiogram (ECG) signals. In particular, we look at the problem of analyzing ST changes in ECG. Changes in the ECG ST segment may reflect transient myocardial ischemia (reduced blood flow to the heart) that can damage heart muscle.
ST changes can also occur due to a variety of other reasons, including changes in heart rate, signal noise, or changes in a person’s temperature or position.
Determining whether ST changes in ECG are due to myocardial ischemia or other causes is a challenging problem. We will train models using data from the PhysioNet Long-Term ST Database to attempt to address this problem. The workshop will take place using python, jupyter notebooks and PyTorch.
Building a High Resolution Clinical Database
Tobias Merz (ADHB), Alex Kazemi (CMDHB), Shawn Sturland (MIT), Leo Anthony Celi (MIT)
This workshop will look at the issues around building, maintaining and using a high resolution health database in clinical practice. The panel have specific experience in the setting of high res ICU data derived from electronic health records. We will discuss the practicalities behind setting up such a database as well as the benefits in its application to daily healthcare. We will also examine barriers to implementation including financial, ethical, structural and data integrity and practical solutions for the NZ healthcare system
The format will consist of a presentation followed by a panel discussion and then a workshop session to look at creating an array of solutions that can be implemented within your healthcare setting.
AI for Medical Image Analysis
Mengling Feng (National University Singapore) and and Caleb Moses (Dragonfly Data Science)
Location: OGGB 4
Artificial intelligence (AI) is the buzzword for all things relating to technology these days. In particular, healthcare is seen as an area in which AI may be gainfully deployed to improve medical care, especially with big data, exponential computing power and a burgeoning demand on healthcare systems due to aging populations. In this talk, I will share two use cases that my group has engaged recently.
Our use case focuses on the analysis of medical images, in particular, the mammogram images. My team recently participated the DREAM digital Mammogram challenged funded by the US white house fund and FDA. The aim was to improve the predictive accuracy of digital mammography for the early detection of breast cancer. The challenge was constructed around 640,000 mammogram images for over 80,000 patients. We developed a patch-based deep neural network structure to extract abnormal lesions and patterns from mammograms so to detect breast cancer. After months of training, the proposed AI model managed to achieve detection performance close to human experts. We are now working with the Breast Cancer Screening Center at National University Hospital aiming to develop and deploy an AI-aided tool to assist radiologists to achieve more accurate, consistent and faster breast cancer screening. Our workshop will share the fundamentals on how deep neural networks can be applied on medical image analysis, and I will also share the lessons learned via our mammogram project. Only pen and paper will be required.