Joint Cardiac Structure Segmentation and Clinical Index Estimation via Multi-Task Learning
DOI: https://doi.org/10.62517/jbdc.202601408
Author(s)
Yali Wan1,*, Deyi Xiong2
Affiliation(s)
1School of Information Technology and Engineering, Guangzhou College of Commerce, Guangzhou, China
2Chongqing Academy of Metrology and Quality Inspection, Chongqing, China
*Corresponding author
Abstract
Cardiac cine-MRI assessment relies on both the segmentation of the ventricular structures and the clinical indices derived from it, most importantly left-ventricular end-diastolic volume (EDV), end-systolic volume (ESV), and ejection fraction (EF). Existing deep-learning methods train a segmentation network and post-hoc compute the indices via Simpson's rule, treating clinical numbers as downstream consequences of the segmentation. We propose JointSegNet, a multi-task network that simultaneously outputs (i) the 4-class segmentation map and (ii) the three clinical indices (EDV, ESV, EF), trained with an uncertainty-weighted compound loss. On ACDC 3-fold cross-validation the proposed JointSegNet achieves a mean foreground Dice of 0.823 ± 0.003 alongside EF MAE of 11.66 ± 1.75% via direct regression (against 17.49 ± 1.27% for a constant-mean reference) and 3.47 ± 1.14% via Simpson’s rule applied to the model’s own segmentations (MAE_EDV 5.7 ± 1.2 mL, MAE_ESV 6.2 ± 0.8 mL). The multi-task setup requires no architectural changes beyond a small regression head, and the uncertainty-weighted loss automatically balances the two objectives.
Keywords
Cardiac Cine-MRI; ACDC; Multi-Task Learning; Clinical Index Estimation; Uncertainty Weighting
References
[1] Lang R M, Badano L P, Mor-Avi V, et al. Recommendations for Cardiac Chamber Quantification by Echocardiography in Adults: An Update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. Journal of the American Society of Echocardiography, 2015, 28(1):1-39.e14. DOI:10.1016/j.echo.2014.10.003.
[2] Bernard O, Lalande A, Zotti C, et al. Deep Learning Techniques for Automatic MRI Cardiac Multi-structures Segmentation and Diagnosis: Is the Problem Solved? IEEE Transactions on Medical Imaging, 2018, 37(11):2514-2525. DOI:10.1109/TMI.2018.2837502.
[3] Isensee F, Jaeger P F, Kohl S A A, et al. nnU-Net: a self-configuring method for deep learning-based biomedical image segmentation. Nature Methods, 2021, 18:203-211. DOI: 10.1038/s41592-020-01008-z.
[4] Ronneberger O, Fischer P, Brox T. U-Net: Convolutional Networks for Biomedical Image Segmentation. Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2015, LNCS 9351:234-241. DOI: 10.1007/978-3-319-24574-4_28.
[5] Oktay O, Schlemper J, Folgoc L L, et al. Attention U-Net: Learning Where to Look for the Pancreas. arXiv: 1804.03999, 2018. DOI:10.48550/arXiv.1804.03999
[6] Cao H, Wang Y, Chen J, et al. Swin-Unet: Unet-like Pure Transformer for Medical Image Segmentation. European Conference on Computer Vision Workshops (ECCVW), 2022, LNCS 13803:205-218. DOI: 10.1007/978-3-031-25066-8_9.
[7] Kendall A, Gal Y, Cipolla R. Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics. IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2018:7482-7491. DOI:10.1109/CVPR.2018.00781.
[8] Caruana R. Multitask Learning. Machine Learning, 1997, 28(1):41-75. DOI: 10.1023/A:1007379606734.
[9] Chen Z, Badrinarayanan V, Lee C Y, Rabinovich A. GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks. International Conference on Machine Learning (ICML), 2018:794-803.
[10] Xue W, Lum A, Mercado A, et al. Full Quantification of Left Ventricle via Deep Multitask Learning Network Respecting Intra- and Inter-task Relatedness. Medical Image Computing and Computer-Assisted Intervention (MICCAI), 2017, LNCS 10435:276-284. DOI: 10.1007/978-3-319-66179-7_32.
[11] Ferdian E, Suinesiaputra A, Fung K, et al. Fully Automated Myocardial Strain Estimation from Cardiovascular MRI-tagged Images Using a Deep Learning Framework in the UK Biobank. Radiology: Cardiothoracic Imaging, 2020, 2(1):e190032. DOI:10.1148/ryct.2020190032.
[12] Grothues F, Smith G C, Moon J C C, et al. Comparison of interstudy reproducibility of cardiovascular magnetic resonance with two-dimensional echocardiography in normal subjects and in patients with heart failure or left ventricular hypertrophy. American Journal of Cardiology, 2002, 90(1):29-34. DOI: 10.1016/S0002-9149(02)02381-0.