CS 523: Research Seminar in Computer Vision and Healthcare

Stanford University   •   Spring 2020-2021   •   11:30am-12:30pm PST

Instructor: Julia Gong

Course Description

[see on Stanford ExploreCourses] With advances in deep learning, computer vision (CV) has been transforming healthcare, from diagnosis to prognosis, from treatment to prevention. Its far-reaching applications include surgical assistants, patient monitoring, data synthesis, and cancer screening. Before these algorithms make their way into the clinic, however, there is exciting research to develop methods that are accurate, robust, interpretable, grounded, and human-centered. In this seminar, we deeply examine these themes in medical CV research through weekly intimate discussions with researchers from academia and industry labs who conduct research at the center of CV and healthcare. Each week students will read and prepare questions and reflections on an assigned paper authored by that week's speaker. We highly encourage students who are interested in taking an interactive, deep dive into medical CV research literature to apply. While there are no hard requirements, we strongly suggest having the background and fluency necessary to read and analyze AI research papers (thus MATH 51 or linear algebra, and at least one of CS 231x, 224x, 221, 229, 230, 234, 238, AI research experience for CV and AI fundamentals may be helpful).

Syllabus

Welcome to CS 523! I'm looking forward to a fantastic quarter ahead, where we'll be reading and discussing medical computer vision papers together. This course aims to bring together enthusiasts and interested students in an intimate setting to deeply discuss medical computer vision research. As a result, an active presence in each course session and completion of the required assignments is required to pass the class. I value open communication, so if you have special circumstances, of course please let me know! :)

Grading. Completing all required assignments + active participation during the class = a passing grade. I'm here to help facilitate learning, not to throw assignments at you - so I'm keeping assignments to a minimum to optimize for this.

Unless otherwise specified, the assignments due before each class session are the following:
(1) Carefully read this week's assigned paper written by this week's speaker. Be prepared to discuss with the class and the speaker!
(2) By 5:00pm PST the day before, post to Canvas at least 4 pre-discussion points regarding the paper; these must be from at least two different categories below.

  • A meaty (not just yes/no) question,
  • An insight or analytical comment about the paper,
  • A key finding or contribution of the paper,
  • A concern or constructive comment about the paper,
  • A related paper it reminds you of that you want to share (and why).

(3) Submit your thank-you note to the post-session thank-you form for the speaker by 5:00pm PST the same day. It only needs to be a couple sentences; in it, make sure you mention one specific insight or takeaway you learned from talking with the speaker.

Date Paper + Announcements Speaker(s)
03.31 Bedside Computer Vision — Moving Artificial Intelligence from Driver Assistance to Patient Safety. New England Journal of Medicine 2018. [link]
A Computer Vision System for Deep Learning-Based Detection of Patient Mobilization Activities in the ICU. Nature Partner Journals (NPJ) Digital Medicine 2019. [link]
Automatic Detection of Hand Hygiene Using Computer Vision Technology. Journal of the American Medical Informatics Association (JAMIA) 2020. [link]
Holistic 3D Human and Scene Mesh Estimation from Single View Images. CVPR 2021. [link]
Presenters: Serena Yeung, Stanford University (MARVL);
Zhenzhen (Jen) Weng, Stanford University;
F. Christopher Holsinger, Stanford Medicine
04.07 Dense Depth Estimation in Monocular Endoscopy With Self-Supervised Learning Methods. IEEE Transactions on Medical Imaging 2020. [link] Presenter: Xingtong Liu, Johns Hopkins University
Visiting Advisor: Mathias Unberath (ARCADE)
04.14 Objective Assessment of Intraoperative Technical Skill in Capsulorhexis using Videos of Cataract Surgery. IJCARS 2019. [link]
Encouraged further reading (data collection paper): Crowdsourcing Annotation of Surgical Instruments in Videos of Cataract Surgery. LABELS 2018, CVII 2018, STENT 2018. [link]
Encouraged further reading (phase identification paper): Assessment of Automated Identification of Phases in Videos of Cataract Surgery Using Machine Learning and Deep Learning Techniques. JAMA Network Open 2019. [link]
Presenter: Tae Soo Kim, Johns Hopkins University
Visiting Advisor: Gregory Hager (CIRL)
04.21 Disentangling Human Error from the Ground Truth in Segmentation of Medical Images. NeurIPS 2020. [link]
Encouraged further reading: Learning From Noisy Labels by Regularized Estimation Of Annotator Confusion. CVPR 2019. [link]
Presenter: Ryutaro Tanno, Microsoft Research Cambridge
04.28 Cortical Response to Naturalistic Stimuli is Largely Predictable with Deep Neural Networks. Preprint 2020. [link] Presenter: Meenakshi Khosla, Cornell University
Visiting Advisor: Mert Sabuncu (Sabuncu Lab)
05.05 TBD Presenter: Adrian Dalca, A.A. Martinos Center for Biomedical Imaging; Massachusetts General Hospital, Harvard Medical School; Massachusetts Institute of Technology Computer Science and Artificial Intelligence Lab
05.12 Scanner Invariant Representations for Diffusion MRI Harmonization. Magn Reson Med. 2020. [link] Presenter: Daniel Moyer, Massachusetts Institute of Technology (Golland Lab)
05.19 Improving the Ability of Deep Neural Networks to Use Information from Multiple Views in Breast Cancer Screening. MIDL 2020. [link] Presenter: Nan Wu, New York University
Visiting Advisor: Krzysztof Geras
05.26 Motion2Vec: Semi-Supervised Representation Learning from Surgical Videos. ICRA 2020. [link] Presenter: Ajay Tanwani, Google Research, Verily (Previously at UC Berkeley)
06.02 Deep Learning-Enabled Breast Cancer Hormonal Receptor Status Determination from Base-Level H&E Stains. Nature Communications 2020. [link]

Encouraged further reading: Dermatologist-Level Classification of Skin Cancer with Deep Neural Networks. Nature 2017. [link]

Additional final assignment: write a paragraph reflection on what you took from the course, the paper(s)/speaker(s) you found most interesting and why, anything that surprised you, and optional feedback on the course.
Presenter: Andre Esteva, Salesforce Research