Challenges and solutions of echocardiography generalization for deep learning: a study in patients with constrictive pericarditis

• Published in: Journal of medical imaging (Bellingham, Wash.) Vol. 10; no. 5; p. 054502
• Main Authors: Jeong, Jiwoong, Chao, Chieh-Ju, Arsanjani, Reza, Kim, Kihong, Pelkey, Melissa N., Chen, Yi-Chieh, Ramzan, Raheel N., Elbahnasawy, Mohammad, Sleem, Mohamed, Ayoub, Chadi, Farina, Juan Maria M., Grogan, Martha, Kane, Garvan C., Patel, Bhavik N., Oh, Jae K., Banerjee, Imon
• Format: Journal Article
• Language: English
• Published: United States Society of Photo-Optical Instrumentation Engineers 01-09-2023
• Subjects: Computer-Aided Diagnosis; deep learning; transthoracic echocardiography; restrictive cardiomyopathy
• ISSN: 2329-4302; 2329-4310
• DOI: 10.1117/1.JMI.10.5.054502

The inherent characteristics of transthoracic echocardiography (TTE) images such as low signal-to-noise ratio and acquisition variations can limit the direct use of TTE images in the development and generalization of deep learning models. As such, we propose an innovative automated framework to address the common challenges in the process of echocardiography deep learning model generalization on the challenging task of constrictive pericarditis (CP) and cardiac amyloidosis (CA) differentiation.
Patients with a confirmed diagnosis of CP or CA and normal cases from Mayo Clinic Rochester and Arizona were identified to extract baseline demographics and the apical 4 chamber view from TTE studies. We proposed an innovative preprocessing and image generalization framework to process the images for training the ResNet50, ResNeXt101, and EfficientNetB2 models. Ablation studies were conducted to justify the effect of each proposed processing step in the final classification performance.
The models were initially trained and validated on 720 unique TTE studies from Mayo Rochester and further validated on 225 studies from Mayo Arizona. With our proposed generalization framework, EfficientNetB2 generalized the best with an average area under the curve (AUC) of 0.96 (
) and 0.83 (
) on the Rochester and Arizona test sets, respectively.
Leveraging the proposed generalization techniques, we successfully developed an echocardiography-based deep learning model that can accurately differentiate CP from CA and normal cases and applied the model to images from two sites. The proposed framework can be further extended for the development of echocardiography-based deep learning models.