CS5540: Computational Techniques for Analyzing Clinical Data

Course description and pre-requisites

CS5540 is a masters-level course that covers a wide range of clinical problems and their associated computational challenges. The practice of medicine is filled with digitally accessible information about patients, ranging from EKG readings to MRI images to electronic health records. This poses a huge opportunity for computer tools that make sense out of this data. Computation tools can be used to answer seemingly straightforward questions about a single patient's test results (“Does this patient have a normal heart rhythm?”), or to address vital questions about large populations (“Is there any clinical condition that affects the risks of Alzheimer”). In CS5540 we will look at many of the most important sources of clinical data and discuss the basic computational techniques used for their analysis, ranging in sophistication from current clinical practice to state-of-the-art research projects.

There are no pre-requisites beyond programming skill at the level of CS2110, although some familiarity with elementary statistics and algorithms would be helpful. The course is being taught in conjunction with Ashish Raj from Cornell's Weill Medical College, and several lectures will be given by physicians.

Credit: 3 credits, grade or S/U.

Meeting: WF 1:25–2:40 in 315 Upson Hall.

Course staff and office hours

Professors: Ramin Zabih (4130 Upson Hall; office hours TBA) and Ashish Raj (Weill Cornell Medical College).

TA: Devin Kennedy (office hours: F 2:30–5:00 PM in UP 317, or by appointment).

Topics (tentative)

• Introduction to medical data and signals: examples drawn from hospital records, EEG, CT, MRI.
• Methods for processing of 1D medical signals
• Arrythmia and heart murmur detection from EEG signals
• Statistically optimal estimation and detection of 1D signals
• Methods for processing multi-dimensional medical signals
• Statistically optimal estimation and detection
• MRI image reconstruction
• Image segmentation
• Medical applications of machine learning
• Classification, clustering and segmentation of tumor/non-tumor voxels from contrast-enhanced MRI of liver and brain
• Classification of patients' disease state from imaging and non-imaging data

Grading and course policies

Details are still being finalized, but the course will include two or three programming assignments, to be completed in groups of two. There will also be a final project.

There will be a few short, in-class quizzes; the purpose of these quizzes is primarily to ensure that students are keeping up with lectures, so they should be fairly easy.

Announcements

Announcements will be posted here as they come up. You can also follow the course RSS feed.

1. (7 March) Assignment 1 has been updated and the due date has been changed to 15 March.
2. (13 March) Assignment 1 has been updated again; we've added a new commandline argument to the contract which you may elect to support. Additionally, we've provided some extras which you may find useful.
3. (31 March) You can consult the project list if you are looking for ideas for your final project.

Lectures

1. (27 January) Introduction — AED, drug side effects, epilepsy diagnosis; includes “Computational Techniques Applicable to Medical Data: One Clinician's View” (lecture given by Gary S. Dorfman, M.D.). PDF, PPTX. References and further reading.
2. (29 January) Introduction — Detection, Estimation, Classification. PDF, PPTX. References and further reading.
3. (3 February) General Methods for Analyzing 1D Data — zero-crossings, local averaging, convolution, matched filters. PDF, PPTX. References and further reading.
4. (5 February) Linear Time Invariance — convolution, random variables, limit theorems, linearity, edge detection. PDF, PPTX. References and further reading.
5. (10 February) Lecture 5: Transforms — Fourier and Wavelets. PDF, PPTX. References and further reading.
6. (12 February) Lecture 6: Classification — k-NN, validation, statistical classification. PDF, PPTX. References and further reading.
7. (17 February) Lecture 7: Estimation — Least Squares, Maximum Likelihood. PDF, PPTX. References and further reading.
8. (19 February) Lecture 8: ECG Analysis (Guest Lecturer: Ken Birman). PDF.
9. (24 February) Lecture 9: Estimation continued — Maximum Likelihood and Maximum a Posteriori Estimation. PDF, PPTX. References and further reading.
10. (26 February) Class canceled due to inclement weather.
11. (3 March) Estimation Continued — Maximum a Posteriori Estimation. PDF, PPTX. References and further reading.
12. (5 March) Decisions Based on Densities — Expectation Maximization. PDF, PPTX. References and further reading.
13. (10 March) (pending)
14. (12 March) (pending)
15. (17 March) Graph cuts. PDF, PPTX.
16. (19 March) (pending)
17. (31 March) Graph cuts. PDF, PPTX.
18. (2 April) (pending)
19. (7 April) Graph cuts. PDF, PPTX.
20. (9 April) (pending)
21. (14 April) PDF, PPTX.
22. (16 April) Accelerated MRI reconstruction. PDF, PPTX.
23. (21 April) Automated dose tracking (guest lecture by George Shih and Devin Kennedy). (slides pending)
24. (23 April) Dynamic MRI image reconstruction. PDF, PPTX.

Lecture notes and slides will be posted here as they become available.

Assignments

Assignments will be posted here as they become available. Submissions and grading will be managed through CMS.

1. Project 1: Analyzing and classifying ECGs. Released 17 February (updated on 7 March); due 15 March. Specification, Data (warning: 94MB download). Useful extras: a MATLAB function for plotting annotated signals; sampling frequencies for normal and shockable signals in the provided data.
2. Project 2: Binary segmentation. Released 21 April; due 7 May. Specification, Data, patched GC library.