Comparison of performance of different automated visual evaluation quality assessment modules

Francisca Morgado; Reuven Weiser; Pedro F. M. Dias; Kelwin Fernandes; David Levitz

doi:10.1117/12.2651687

14 March 2023 Comparison of performance of different automated visual evaluation quality assessment modules

Francisca Morgado, Reuven Weiser, Pedro F. M. Dias, Kelwin Fernandes, David Levitz

Author Affiliations +

Proceedings Volume PC12369, Optics and Biophotonics in Low-Resource Settings IX; PC1236906 (2023) https://doi.org/10.1117/12.2651687
Event: SPIE BiOS, 2023, San Francisco, California, United States

Abstract

Background: Cervical cancer disproportionally harms women in low and middle income countries (LMICs). There is increasing interest in automated visual evaluation (AVE) – using artificial intelligence to analyze cervical images at the point of care (PoC) – for managing patients in LMICs. AVE has a diagnostic component (for pathology) and a quality component (to ensure image adequacy). The quality component must run on the imaging device at the PoC, and is limited by its processors for computation. Methods: A novel, multiple-module algorithm for assessing cervical image quality was developed in an Android application. One module located the cervix and another sought objects obstructing the transformation zone. The cervix locator module is an object detection model that determined the bounding box of the cervix. Models trained on multiple architectures (YOLOv5 and EfficientDet-Lite2) with the same data were compared. For obstructions classification, a multi-task model was trained to detect 5 common obstructions (blood, SCJ inside of os, loose vaginal walls, mucus, blur/glare) and obstruction-free cervix. Performance of the model’s tasks were compared for 2 different imaging devices. Results and Discussion: The cervix locator performed better and was faster for YOLOv5, although differences were minimal. In the obstructions classifier, 4 different tasks (loose vaginal walls, blood, SCJ inside of os, and obstruction-free) performed satisfactorily. For all modules, the full computation time was <10 sec. Both modules met the desired performance thresholds for image adequacy assessment. The algorithm shown here is to our knowledge, the first AVE quality classifier running on a mobile device.

Conference Presentation

Citation Download Citation

Francisca Morgado, Reuven Weiser, Pedro F. M. Dias, Kelwin Fernandes, and David Levitz "Comparison of performance of different automated visual evaluation quality assessment modules ", Proc. SPIE PC12369, Optics and Biophotonics in Low-Resource Settings IX, PC1236906 (14 March 2023); https://doi.org/10.1117/12.2651687

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

;

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE