Deep Doc | Issue 1
Deep Doc is a newsletter aimed at bringing the research and news around the intersection of biomedicine, healthcare and AI. This will include different research articles, new or impactful technologies including both software and hardware, policies affecting the space of healthcare or AI and pointers to different events and opportunities.
Although this information is available and accessible easily, we feel that they are scattered and with Deep Doctor, we aim to bring them together, hoping that it will help researchers and readers to navigate easily and eventually contribute to the future research and technology for healthcare and AI. We found some newsletters that are already out there covering the space of healthcare and machine learning. One notable mention goes to Doctor Penguin.
Here are the highlights of this version,
Johnson et al have released MIMIC-IV-ED corpus. MIMIC-ED is a large (and freely available!) dataset containing information about vital signs, triage information, discharge diagnoses, etc of nearly 450K emergency department (ED) stays. The dataset could be significant in understanding the dynamics involving emergency visits and building models to provide better and prompt care.
A new paper from Rojas et al. published in Nature communication have identified six new genetic variants associated with Alzheimer’s Disease (AD). They report their work on the largest Genome Wide Association Studies (GWAS) for AD to date, consisting of genetic information from nearly half-million individuals of European ancestry. As the paper also acknowledges, the generalization of this study to other ancestries or populations remains unanswered.
In recent years the neural network based models have excelled at learning complex relationships from large datasets, but interpreting their predictions remains challenging. A recent preprint by Linder et al from University of Washington presents a new method for interpreting neural networks for biological sequences by learning stochastic masks with rigorous analysis and case studies.
Redesigning AI, a new essay-collection book on artificial intelligence has been published. With that background that the advancement of AI technologies and its promises have been increasingly overshadowed by its perils, including disinformation and bias, the book asks what can be done to redirect AI for the good of everyone. The book has an exciting list of contributors. One notable mention goes to Rachel Thomas’s essay “Medicine’s Machine Learning Problem”.
Fei Fei Li and Andrew Ng discuss the intersection of AI and healthcare in a virtual fireside chat co-hosted by Stanford University’s Institute for Human Centered AI (HAI), pointing out the AI’s value to healthcare, challenges in implementation and concerns around biases. Both of them agreed and advocated for the importance of AI technology in the critical domain of healthcare.
Events:
IMLH 2021 (Interpretable Machine Learning in Healthcare) is happening on Friday, July 23, 2021. This is ICML 2021 workshop that aims to bring together researchers in ML, computer vision, NLP, healthcare, medicine, etc. to facilitate discussions on interpretable medical machine intelligence. Please check their websites for more information.