A novel machine-learning algorithm to estimate the position and size of myocardial infarction for MRI sequence
- 24 Downloads
Early and accurate assessment of myocardial abnormalities is of utmost importance in the diagnosis of myocardial infarction (MI). In this study, we proposed a machine-learning and image motion statistic based approach to automating the detection and localization of MI area in magnetic resonance images. Unlike the existing techniques, the proposed method that could be directly acquired position and size of MI area with sub-pixel precision. Standard clinical magnetic resonance image and delayed enhancement imaging data of 58 patients with MI were used for developing this algorithm. First, we are located and extracted the LV from the original MR image. Then, we using a novel Optical Flow algorithm to building statistical image motion features include the motion trajectories of each Point on myocardium in whole frames. In the end, we using these results as inputs to a support vector machine classifier, which can obtain an MI area assessment of the myocardium. Compared to the pixel by pixel in delayed enhancement imaging, the proposed algorithm yielded the highest classification accuracy of 93.34% and the kappa measure of 0.74. The field experiments our method performs significantly better than other recent methods, and can lead to a promising diagnostic support tool to assists clinicians, particularly for novice readers with limited experience.
KeywordsMyocardial infarction Optical flow Motion feature
Mathematics Subject Classification68T10
Funding was provided by National Key Research and Development Program of China (Grant No. 2016YFC1300300), The Natural Science Foundation of Jiangsu Province of China (Grant BK20150931). National Natural Science Foundation of China (Grant Nos. 61771464, U1801265, 61702275, 41775008, 61673020), Shen Zhen Research and Innovation Funding (Grant No. JCYJ20170307165309009), Provincial Natural Science Research Program of the Higher Education Institutions of Anhui Province (Grant No. KJ2016A016).
- 7.Burton A, Radford J (1978) Thinking in perspective: critical essays in the study of thought processes, vol 646. Routledge, AbingdonGoogle Scholar
- 8.Chang CC, Lin CJ (2011) Libsvm: a library for support vector machines. ACM Trans Intell Syst Technol (TIST) 2(3):27Google Scholar
- 9.Charles R (2000) Applying computational mechanics analysis to biological systems. In: 18th applied aerodynamics conference, p 4338Google Scholar
- 10.Chéhikian A (1992) Optimal algorithms for low-pass and laplacian image pyramids computation. Traitement du Signal 9:297–297Google Scholar
- 14.Hoffmann R, von Bardeleben S, Kasprzak JD, Borges AC, ten Cate F, Firschke C, Lafitte S, Al-Saadi N, Kuntz-Hehner S, Horstick G et al (2006) Analysis of regional left ventricular function by cineventriculography, cardiac magnetic resonance imaging, and unenhanced and contrast-enhanced echocardiography: a multicenter comparison of methods. J Am Coll Cardiol 47(1):121–128CrossRefGoogle Scholar
- 19.Lei C, Yang YH (2009) Optical flow estimation on coarse-to-fine region-trees using discrete optimization. In: 2009 IEEE 12th international conference on computer vision. IEEE, pp 1562–1569Google Scholar
- 20.Leung KE, Bosch JG (2007) Localized shape variations for classifying wall motion in echocardiograms. In: International conference on medical image computing and computer-assisted intervention. Springer, Berlin, pp 52–59Google Scholar
- 22.Mansor S, Noble JA (2008) Local wall motion classification of stress echocardiography using a hidden markov model approach. In: 5th IEEE international symposium on biomedical imaging: from nano to macro, 2008. ISBI 2008. IEEE, pp 1295–1298Google Scholar
- 30.Viera AJ, Garrett JM et al (2005) Understanding interobserver agreement: the kappa statistic. Fam Med 37(5):360–363Google Scholar