PhD Seminar • Machine Learning • Exploring the Impact of Self-Supervised Pretraining for Diagnostic Tasks in Lung UltrasoundExport this event to calendar

Tuesday, August 8, 2023 — 9:30 AM to 10:30 AM EDT

Please note: This PhD seminar will take place in DC 1331.

Blake VanBerlo, PhD candidate
David R. Cheriton School of Computer Science

Supervisors: Professors Jesse Hoey, Alexander Wong

Lung ultrasound (LUS) is an increasingly popular examination in critical care scenarios, owing to its inexpensiveness, portability, safety, and diagnostic accuracy. Given the lack of professionals trained in LUS interpretation, a handful of recent studies have explored the usage of machine learning methods for the automation of LUS diagnostic tasks. Like many domains in medical imaging, access to labelled examples is limited. When unlabelled examples greatly outnumber labelled examples, labelling is a time-consuming and costly barrier to developing diagnostic systems with machine learning.

We explore the application of self-supervised pretraining methods to make use of unlabelled LUS videos. Focusing on joint embedding methods, we examine the impact of initializing deep neural networks with self-supervised pretrained weights on their performance on LUS diagnostic tasks. We investigate a variety of core classification tasks in M-mode and B-mode LUS, such as absent lung sliding detection and pleural effusion classification. Results indicate that self-supervised pretraining reduces reliance on labelled data while improving performance on external datasets.

Location 
DC - William G. Davis Computer Research Centre
DC 1331
200 University Avenue West

Waterloo, ON N2L 3G1
Canada
Event tags 

S M T W T F S
30
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
1
2
3
  1. 2024 (186)
    1. September (1)
    2. August (6)
    3. July (21)
    4. June (17)
    5. May (23)
    6. April (41)
    7. March (27)
    8. February (25)
    9. January (25)
  2. 2023 (296)
    1. December (20)
    2. November (28)
    3. October (15)
    4. September (25)
    5. August (30)
    6. July (30)
    7. June (22)
    8. May (23)
    9. April (32)
    10. March (31)
    11. February (18)
    12. January (22)
  3. 2022 (245)
  4. 2021 (210)
  5. 2020 (217)
  6. 2019 (255)
  7. 2018 (217)
  8. 2017 (36)
  9. 2016 (21)
  10. 2015 (36)
  11. 2014 (33)
  12. 2013 (23)
  13. 2012 (4)
  14. 2011 (1)
  15. 2010 (1)
  16. 2009 (1)
  17. 2008 (1)