A generalized deep learning model for heart failure diagnosis using dynamic and static ultrasound

• DOI: 10.2478/jtim-2023-0088
• Publication Date: 2023-10-31T09:07:21Z
• Authors: Zeye Liu; Yuan Huang; Hang Li; Wenchao Li; Fengwen Zhang; Wenbin Ouyang; Shouzheng Wang; Zhiling Luo; Jinduo Wang; Yan Chen; Ruibing Xia; Yakun Li; Xiangbin Pan
• Keywords: Cardiac Ultrasound

Echocardiography (ECG) is the most common method used to diagnose heart failure (HF). However, its accuracy relies on experience of operator. Additionally, video format data makes it challenging for patients bring them referrals and reexaminations. Therefore, this study a deep learning approach assist physicians in assessing cardiac function promote standardization echocardiographic findings compatibility dynamic static ultrasound data.A spatio-temporal convolutional model r2plus1d-Pan (trained applied data) was improved trained using idea "regression training combined with classification application," which can be generalized ECG views identify HF reduced ejection fraction (EF < 40%). three independent datasets containing 8976 10085 videos were established. Subsequently, multinational, multi-center dataset EF labeled. Furthermore, validation performed. Finally, 15 registered ultrasonographers cardiologists different working years regional hospitals specialized cardiovascular disease recruited compare results.The proposed achieved an area under receiveroperating characteristic curve (AUC) value 0.95 (95% confidence interval [CI]: 0.947 0.953) set AUC 1 CI, 1) set. view (validation set) simultaneous input 1, 2, 4, 8 images same heart, accuracies 85%, 81%, 93%, 92%, respectively. On data, artificial intelligence (AI) comparable best performance more than 3 (P = 0.344), but significantly better median level 0.0000008).A new convolution constructed accurately (< 40%) images. The outperformed diagnostic senior specialists. This may first HF-related AI compatible multi-dimensional thereby contribute improvement diagnosis. enables carry "on-the-go" reports referral reexamination, thus saving healthcare resources.