We present a procedure that detects myocardial infarction by analyzing left ventricular shapes recorded at end-diastole and endsystole, involving both shape and statistical analyses. In the framework of Geometric Morphometrics, we use Generalized Procrustes Analysis, and optionally an Euclidean Parallel Transport, followed by Principal Components Analysis to analyze the shapes. We then test the performances of different classification methods on the dataset. Among the different datasets and classification methods used, we found that the best classification performance is given by the following workflow: full shape (epicardium+endocardium) analyzed in the Shape Space (i.e. by scaling shapes at unit size); successive Parallel Transport centered toward the Grand Mean, in order to detect pure deformations; final statistical analysis via Support Vector Machine with radial basis Gaussian function. Healthy individuals show both a stronger contraction and a shape difference in systole with respect to pathological subjects. Moreover, endocardium clearly presents a larger deformation when contrasted with epicardium. Eventually, the solution for the blind test dataset is given. When using Support Vector Machine for learning from the whole training dataset and for successively classifying the 200 blind test dataset, we obtained 96 subjects classified as normal and 104 classified as pathological. After the disclosure of the blind dataset this resulted in 95% of total accurracy with sensitivity at 97% and specificity at 93%.

Piras, P., Teresi, L., Gabriele, S., Evangelista, A., Esposito, G., Varano, V., et al. (2016). Systo-diastolic LV shape analysis by geometric morphometrics and parallel transport highly discriminates myocardial infarction. In Statistical Atlases and Computational Models of the Heart (pp. 119-129) [10.1007/978-3-319-28712-6_13].

Systo-diastolic LV shape analysis by geometric morphometrics and parallel transport highly discriminates myocardial infarction

TERESI, Luciano;GABRIELE, STEFANO;VARANO, VALERIO;
2016

Abstract

We present a procedure that detects myocardial infarction by analyzing left ventricular shapes recorded at end-diastole and endsystole, involving both shape and statistical analyses. In the framework of Geometric Morphometrics, we use Generalized Procrustes Analysis, and optionally an Euclidean Parallel Transport, followed by Principal Components Analysis to analyze the shapes. We then test the performances of different classification methods on the dataset. Among the different datasets and classification methods used, we found that the best classification performance is given by the following workflow: full shape (epicardium+endocardium) analyzed in the Shape Space (i.e. by scaling shapes at unit size); successive Parallel Transport centered toward the Grand Mean, in order to detect pure deformations; final statistical analysis via Support Vector Machine with radial basis Gaussian function. Healthy individuals show both a stronger contraction and a shape difference in systole with respect to pathological subjects. Moreover, endocardium clearly presents a larger deformation when contrasted with epicardium. Eventually, the solution for the blind test dataset is given. When using Support Vector Machine for learning from the whole training dataset and for successively classifying the 200 blind test dataset, we obtained 96 subjects classified as normal and 104 classified as pathological. After the disclosure of the blind dataset this resulted in 95% of total accurracy with sensitivity at 97% and specificity at 93%.
Piras, P., Teresi, L., Gabriele, S., Evangelista, A., Esposito, G., Varano, V., et al. (2016). Systo-diastolic LV shape analysis by geometric morphometrics and parallel transport highly discriminates myocardial infarction. In Statistical Atlases and Computational Models of the Heart (pp. 119-129) [10.1007/978-3-319-28712-6_13].
File in questo prodotto:
Non ci sono file associati a questo prodotto.

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: http://hdl.handle.net/11590/310093
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 9
  • ???jsp.display-item.citation.isi??? ND
social impact