Cardiac magnetic resonance (CMR) represents the gold standard for the diagnosis of cardiovascular diseases. We developed a deep learning approach for the automatic detection and segmentation of left and right ventricles and myocardium (Myo) on short-axis cine CMR images, including all clinically relevant slices. A dataset of 210 studies (3 pathology groups) was considered: Images were acquired and manually segmented (gold standard, GS) at Centro Cardiologico Monzino (Milan, Italy). Automatic segmentation was performed with a U-Net inspired architecture were two loss function were used: weighted cross entropy (WCE) and its combination with the Dice loss function (WCE+Dice). Two experiments were conducted: A) all the slices were included; ii) slices where the Myo did not completely surrounded the LV were removed. To evaluate the clinical relevance of our approach, the predicted segmentation was reviewed and corrected by an expert physician. The two loss function performed similarly, with slightly better results for WCE, resulting in a strong correlation with the manually-adjusted segmentation.

Penso, M., Moccia, S., Scafuri, S., Muscogiuri, G., Pontone, G., Pepi, M., et al. (2020). Automated Left and Right Chamber Segmentation in Cardiac MRI Using Dense Fully Convolutional Neural Network. In Computing in Cardiology. IEEE Computer Society [10.22489/CinC.2020.247].

Automated Left and Right Chamber Segmentation in Cardiac MRI Using Dense Fully Convolutional Neural Network

Muscogiuri G.;
2020

Abstract

Cardiac magnetic resonance (CMR) represents the gold standard for the diagnosis of cardiovascular diseases. We developed a deep learning approach for the automatic detection and segmentation of left and right ventricles and myocardium (Myo) on short-axis cine CMR images, including all clinically relevant slices. A dataset of 210 studies (3 pathology groups) was considered: Images were acquired and manually segmented (gold standard, GS) at Centro Cardiologico Monzino (Milan, Italy). Automatic segmentation was performed with a U-Net inspired architecture were two loss function were used: weighted cross entropy (WCE) and its combination with the Dice loss function (WCE+Dice). Two experiments were conducted: A) all the slices were included; ii) slices where the Myo did not completely surrounded the LV were removed. To evaluate the clinical relevance of our approach, the predicted segmentation was reviewed and corrected by an expert physician. The two loss function performed similarly, with slightly better results for WCE, resulting in a strong correlation with the manually-adjusted segmentation.
No
paper
Cardiology; Convolutional neural networks; Deep learning; Diagnosis; Magnetic resonance;
English
2020 Computing in Cardiology, CinC 2020 - 13 September 2020 through 16 September 2020
978-172817382-5
Penso, M., Moccia, S., Scafuri, S., Muscogiuri, G., Pontone, G., Pepi, M., et al. (2020). Automated Left and Right Chamber Segmentation in Cardiac MRI Using Dense Fully Convolutional Neural Network. In Computing in Cardiology. IEEE Computer Society [10.22489/CinC.2020.247].
Penso, M; Moccia, S; Scafuri, S; Muscogiuri, G; Pontone, G; Pepi, M; Caiani, E
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: https://hdl.handle.net/10281/378698
Citazioni
  • Scopus 0
  • ???jsp.display-item.citation.isi??? 0
Social impact