Deep-learning-based quantification of M-Mode images Deep-learning-based quantification of M-Mode images

Doppler echocardiography is a vital imaging modality for the assessment of various cardiac pathologies. In the case of fetuses with compromised cardiovascular systems it has been demonstrated to be a vital diagnostic and monitoring tool through the usage of various modalities (M-Mode, tissue velocity, blood velocity). The pipeline for analysing fetal images is, however, very reliant on manual quantification steps and labour-intensive, pivoting towards the markup of a series of control points on the image and the definition of the envelope of the structure for their posterior analysis. Given the rise and consolidation of deep learning (DL) based solutions on many industrial settings, the application of DL on fetal echocardiographic data is starting to be explored. However, DL-based models are oftentimes regarded as black-box models that provide an output while hiding the decision process. This can have detrimental side effects such as bias in the decision process or due to processing data outside of the inherent variability of the training data pool.

 

In this context, the student's work will focus on direct quantification of the movement of different anatomical structures through envelope extraction from M-Mode data and for its posterior aggregation into its important fiducials for landmark localization. This will be performed by the exploration of different standard DL architectures for segmentation as well as the exploration of more state-of-the-art propositions for enhancing model performance.

 

 

References

[1] Garcia-Canadilla, P., Sanchez-Martinez, S., Crispi, F., & Bijnens, B. (2020). Machine learning in fetal cardiology: What to expect. Fetal diagnosis and therapy, 47(5), 363-372.

[2] Elwazir, M. Y., Akkus, Z., Oguz, D., Ye, Z., & Oh, J. K. (2020, October). Fully Automated Mitral Inflow Doppler Analysis Using Deep Learning. In 2020 IEEE 20th International Conference on Bioinformatics and Bioengineering (BIBE) (pp. 691-696). IEEE.

[3] Jahren, T. S., Steen, E. N., Aase, S. A., & Solberg, A. H. S. (2020). Estimation of End-Diastole in Cardiac Spectral Doppler Using Deep Learning. IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control, 67(12), 2605-2614.

[4] Ronneberger, O., Fischer, P., & Brox, T. (2015, October). U-net: Convolutional networks for biomedical image segmentation. In International Conference on Medical image computing and computer-assisted intervention (pp. 234-241). Springer, Cham.

[5] Ashley, E. A., & Niebauer, J. (2004). Cardiology explained.