Cardiac Landmark Detection using Generative Adversarial Networks from Cardiac MR Images
Aparna Kanakatte (TCS),* DIVYA M BHATIA (TCS), Pavan Kumar Reddy K (TCS Research), Jayavardhana Gubbi (TCS Research), Avik Ghose (TCS)
The 34th British Machine Vision Conference
Abstract
Anatomical landmarks are very important for the structural and functional analysis of the heart. Cardiac magnetic resonance (CMR) images have advanced to become a powerful non-invasive diagnostic tool in clinical practice. The first step in many medical imaging applications is to detect anatomical landmarks accurately. The manual identification of these landmarks is difficult due to their shape and appearance variations across populations, making it time-consuming and operator dependent. We present a GAN-based landmark detection network that can detect smaller objects including landmarks with greater accuracy across varied sample sizes using a proposed modified loss function. The proposed method outperforms other methods reported in literature when trained and tested on the STACOM LV landmark detection challenge dataset. This improved performance is achieved by leveraging the power of the GAN architecture to learn more complex features of the objects being detected. The robustness of the proposed approach is demonstrated by obtaining reduced mean error when blind tested on ACDC dataset.Video
Citation
@inproceedings{Kanakatte_2023_BMVC,
author = {Aparna Kanakatte and DIVYA M BHATIA and Pavan Kumar Reddy K and Jayavardhana Gubbi and Avik Ghose},
title = {Cardiac Landmark Detection using Generative Adversarial Networks from Cardiac MR Images},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year = {2023},
url = {https://papers.bmvc2023.org/0204.pdf}
}
Copyright © 2023
The British Machine Vision Conference is organised by