Deep Learning in Medicine 

BMSC-GA 4493, BMIN-GA 3007

Course Overview

The use of deep networks has revolutionized areas of image recognition, speech recognition, and natural language processing. Deep networks are also transforming the world of medicine by helping doctors to improve detection, diagnosis, treatment, and management of disease. Moreover, researchers begin to incorporate deep learning methods in areas such as drug development, diagnostic radiology and personalized medicine. In this course, we will focus on the deep learning approaches that are practical and currently used in various medical disciplines. In the labs, we will implement the approaches learned in the course using available multi-modal medical dataset.

Learning Objectives

Students will learn the most common deep learning methods emerging in medicine. Students will be able to differentiate various deep learning methods and choose the most appropriate ones for specific research problems.

General Information

Meeting Time and Place

Tuesdays 3:00pm-5:00pm and Thursdays 4:00pm-5:00pm

Location: The Science Building, 435 East 30th St, Ground Floor, Room G19 (435 East 30th St, New York, NY, 10016)

Course Directors

Cem M. Deniz (cem.deniz@nyumc.org)

Narges Razavian (narges.razavian@nyumc.org)

Teaching Assistant

Juiting Hsu (juiting.hsu@nyu.edu)

Jingyi Su (js5991@nyu.edu)

Lee Tanenbaum (leedtan@gmail.com)

TA Office Hours (will be held before or after the Lab sections on Thursdays)

Time/Day

02/08

02/15

02/22

03/01

03/08

03/22

03/29

04/05

04/12

04/19

04/26

05/03

3-4pm

JS

JS

JH

LT

JH

JS

JH

JH

5-6pm

JS

LT

LT

LT

Q&A Site

We will be using Piazza for class discussion. The system is highly catered to getting you help fast and efficiently from classmates, the TA, and us. Rather than emailing questions to the teaching staff, I encourage you to post your questions on Piazza.

Find our class page at: https://piazza.com/nyumc.org/spring2018/bmscga4493

Lab and HW Github Repository

https://github.com/nyumc-dl/BMSC-GA-4493-Spring2018 

Tools for Success

Python 3 is preferred, Jupyter Notebook, PyTorch 

Course Materials

Prerequisites

Introduction to Programming (BMSC-GA 1358) or equivalent

Python Programming Language

Course Assessment

Homeworks (40%) + Final Project (50%: 10% for proposal and final presentation, 30% for the paper) + Attendance (10%)

Auditing (updated on 02/05)

It is possible to audit the course for the NYU community. However, due to current classroom size restrictions, we can first allow registered students to attend the course. We are trying to find a larger classroom, if this happens, audits will be able attend the course without such a restriction. 

Schedule

Day

Topic

Reading / Watching List (WiP)

01/23

Course Overview (CD, NR)

(Lecture 1)

  • Deep Learning, Goodfellow et al, Chapter 2 and Chapter 3.
  • LeCun Y, Bengio Y, Hinton G. Deep learning. Nature. 2015;521(7553):436-444. (pdf)
  • Mukherjee S. AI versus MD What happens when diagnosis is automated? The New Yorker Annals of Medicine. 2017. (link)

01/25

Lab 1: PyTorch and Packages Setup (JH)

01/30

Deep Learning Basics (NR)

(Lecture 2)

  • Deep Learning, Goodfellow et al, Chapter 5.
  • Deo RC, Machine Learning in Medicine, Circulation. 2015;132:1920-30 (pdf)
  • Demis Hassabis et.al. “Neuroscience-Inspired Artificial Intelligence” Neuron. 2017.(link)

02/01

Lab 2: PyTorch Tutorial (JS)

02/06

Deep Feedforward Networks (NR) (Lecture 3)

  • Deep Learning, Goodfellow et al, Chapter 6.
  • Ravi D et al. Deep Learning for Health Informatics. IEEE J Biomed Health Informatics. 2017;21(1):4-21. (pdf)

02/08

HPC Workshop

02/13

Convolutional Networks I (CD) (Lecture 4)

  • Deep Learning, Goodfellow et al, Chapter 9.
  • Litjens G et al. A Survey on Deep Medical Image Analysis. 2017; ArXiv: 1702.05747. (pdf)

02/15

Lab 3: Deep Networks (LT)

02/20

Student Project Proposal Presentations (one page papers are due at 3:00 pm)

02/22

Feedback to Student Project Proposals

02/27

Convolutional Networks II (CD)

(Lecture 5)

  • Deep Learning, Goodfellow et al, Chapter 9.
  • Angermueller C et al. Deep Learning for Computational Biology. Molecular Systems Biology 2016; 12:878. (pdf)
  • 2D Visualization of a CNN (link)

03/01

Lab 4: CNNs (LT)

03/06

Optimization of Deep Networks (CD) (Lecture 6)

  • Deep Learning, Goodfellow et al, Chapter 8.
  • Mamoshina P et al. Applications of Deep Learning in Biomedicine. Mol. Pharmaceutics 2016; 13(5):1445-54. (pdf)

03/08

Guest Lecture (Krzysztof J. Geras)

03/13

03/15

No Class (Spring Break)

03/20

Regularization (NR)

(Lecture 7)

(Lecture 7.5)

  • Deep Learning, Goodfellow et al, Chapter 7.
  • Srivastava N et al, Dropout: A Simple Way to Prevent Neural Networks from Overfitting. Journal of Machine Learning Research 2014; 15: 1929:1958. (pdf)

03/22

Lab: Regularization (JS)

03/27

Practical Methodology and Applications (CD) (Lecture 8)

  • Deep Learning, Goodfellow et al, Chapter 11 and Chapter 12.
  • Bengio Y. Practical Recommendations for Gradient-Based Training of Deep Architectures. In: Montavon G, Orr GB, Müller K-R, eds. Neural Networks: Tricks of the Trade: Second Edition. Berlin, Heidelberg: Springer Berlin Heidelberg; 2012:437-478 (pdf)

03/29

Lab:  (LT)

04/03

Recurrent and Recursive Nets (NR) (Lecture 9)

  • Deep Learning, Goodfellow et al, Chapter 10.
  • Zachary C et al. Learning to Diagnose with LSTM Recurrent Neural Networks. 2016; arXiv: 1511.03677 (pdf)

04/05

Lab: RNNs (JH)

04/10

Generative Adversarial Networks (NR)

(Lecture 10)

  • Deep Learning, Goodfellow et al, Chapter 15.
  • Goodfellow IJ et al. Generative Adversarial Networks. 2014; arXiv:1406.2661 (pdf)

Binge Watch

  • NIPS 2016 Workshop on Adversarial Training - Ian Goodfellow - Introduction to GANs (link)

04/12

Lab: GANs (JS)

04/17

Autoencoders (CD) (Lecture 11)

  • Deep Learning, Goodfellow et al, Chapter 14.
  • Miotto R et al. Deep Patient: An Unsupervised Representation to Predict the Future of Patients from Electronic Health Records. Scientific Reports 6;26094. (pdf)
  • Ronneberger O et al. U-Net: Convolutional Networks for Biomedical Image Segmentation 2015; arXiv:1505.0459 (pdf)

04/19

Lab: Autoencoders (JH)

04/24

Transfer Learning (CD, NR) (Lecture 12)

  • Deep Learning, Goodfellow et al, Chapter 15.
  • Esteva A, Kuprel B, Novoa RA, et al. Dermatologist-level classification of skin cancer with deep neural networks. Nature. 2017;542(7639). (pdf)

04/26

Lab: Transfer Learning (LT)

05/01

No Class (Finalize your projects)

05/03

No Lab ( Finalize your projects)

05/07

Student Project Paper Deadline (3:00 pm)

05/08

Student Project Presentations

Homeworks

There will be four homeworks each will be graded for 10% of your final grade.

HW1 (due 02/13 3:00 pm): Regression related to medical field - getting familiar with deep learning tools

HW2 (due 03/16 5:00 pm): Basics of convnets - finding abnormalities on a medical dataset

HW3 (due 04/10/2018 3:00 pm): Sequence Classification¶

HW4 (due 05/12/2018 5:00 pm): Transfer learning

Student Projects

Students are expected to work on a project related to the topics presented in the lectures.

Teams

You may work in teams of up to four on the final project. Teams must be finalized by the project proposal deadline (02/20).

Topics

Students are free to choose their project topics related to medicine. Example research topics are:

NYU internal (if you are interested in one of these topics and learn more about them, please send an email to cem.deniz@nyumc.org so I can share the contact information with your team)

External

Proposal

You must submit a your one page proposal and make a presentation to receive useful feedback on the project.

Paper and Presentation

The final report should include the description of the task, models, experiments and conclusion. It can be up to 6 pages long excluding unlimited pages reserved for references. In the final week of the semester you will present your findings in the class. Both of your paper and presentation will affect your grade on the student project.

Contribution Statements

The final report must state the contributions of each team member. Team projects which fail to include this will receive a 2.5% grade deduction.

General Policies

Late/missed work

You must adhere to the due dates for all required submissions. If you miss a deadline, then you will not get credit for that assignment/post.

Incomplete

No "Incomplete" will be assigned for this course unless we are at the very end of the course and you have an emergency.

Responding to Messages

We use piazza for all course related questions/discussions related to the course. If you have highly private/sensitive issues and prefer to communicate via email, please keep in mind that: we will check emails daily during the week, and we will respond to you within 48 hours.

Announcements

We will make announcements throughout the semester by e-mail.

Make sure that your email address is updated; otherwise you may miss important emails from me.

Safeguards

Always backup your work on a safe place (electronic file with a backup is recommended) and make a hard copy. Do not wait for the last minute to do your work. Allow time for deadlines.

Plagiarism

Plagiarism, the presentation of someone else's words or ideas as your own, is a serious offense and will not be tolerated in this class. The first time you plagiarize someone else's work, you will receive a zero for that assignment. The second time you plagiarize, you will fail the course with a notation of academic dishonesty on your official record.