Date and time

MIT E25-111 and Zoom (see below for full information)

Domain and User-Centered Machine Learning for Medical Image Analysis

Machine learning-based (ML) algorithms have gained substantial attention owing to their ability to automatically recognize subtle patterns in medical images. Their potential to augment clinical decision-making addresses the high clinical need for automating cognitively challenging tasks, such as analyzing medical images. These developments will lighten the burden on radiologists and avoid a further increase in healthcare expenditure. However, despite the high enthusiasm for ML algorithms, concerns regarding the readiness for clinical deployment are impeding their clinical translation. Here, we address two fundamental challenges to translating ML algorithms into clinical care settings.

First, the data modeling approach must be carefully selected depending on the specific task. We illustrate the advantages of shifting from a strictly discrete (ordinal) model of disease severity distribution to a continuously valued one. We introduce a generalized framework to recover information lost by discretizing continuous variables into discrete training labels. Additionally, we present the first conception and demonstration of two methods that enable the joint learning of annotators’ ordinal classification and their individual biases for a latent, continuously valued target variable like disease severity.

Second, the performance of ML algorithms needs to be evaluated in a clinically meaningful manner. We address the disconnect between the subjective quality perception of clinical experts and the metrics typically used to assess performance. Furthermore, we identify criteria that experts use to evaluate the quality of automatically generated tumor outlines and describe their thought processes as they correct them.

Based on the learnings from our work, we conclude with concrete recommendations for developing stable ML tools for medical imaging.

Thesis Supervisor:
Jayashree Kalpathy-Cramer, PhD, PhD
Professor of Ophthalmology, Chief of Division of Artificial Medical Intelligence in Ophthalmology, University of Colorado School of Medicine; Visiting Professor in Radiology, Harvard Medical School

Thesis Committee Chair:
Polina Golland, PhD
Sunlin (1966) and Priscilla Chou Professor of Electrical Engineering and Computer Science, MIT Computer Science and Artificial Intelligence Lab, Affiliate Faculty, MIT Institute for Medical Engineering & Science

Thesis Readers:
Bruce Fischl, PhD
Professor in Radiology, Harvard Medical School

Clifton D. Fuller, MD, PhD
Professor, Division of Radiation Oncology, The University of Texas MD Anderson Cancer Center

Zoom invitation – 

Katharina Hoebel is inviting you to a scheduled Zoom meeting.

Topic: Kathi Hoebel's PhD Thesis Defense

Time: Thursday, January 26, 2023 2:00 PM Eastern Time (US and Canada)

Your participation is important to us: please notify hst [at], at least 3 business days in advance, if you require accommodations in order to access this event.

Join Zoom meeting

Password: 920996

Join by telephone (use any number to dial in)

        +1 646 931 3860

        +1 929 436 2866

        +1 301 715 8592

        +1 305 224 1968

        +1 309 205 3325

        +1 312 626 6799

        +1 360 209 5623

        +1 386 347 5053

        +1 507 473 4847

        +1 564 217 2000

        +1 669 444 9171

        +1 669 900 6833

        +1 689 278 1000

        +1 719 359 4580

        +1 253 205 0468

        +1 253 215 8782

        +1 346 248 7799

International numbers available:

One tap mobile: +16469313860,,93324909620# US

    Join by SIP conference room system

Meeting ID: 933 2490 9620

93324909620.920996 [at]