Echo-SyncNet: Self-Supervised Cardiac View Synchronization in Echocardiography

In echocardiography (echo), an electrocardiogram (ECG) is conventionally used to temporally align different cardiac views for assessing critical measurements. However, in emergencies or point-of-care situations, acquiring ECG often not option, hence motivating the need alternative temporal synchronization methods. Here, we propose Echo-SyncNet, a self-supervised learning framework synchronize various cross-sectional 2D echo series without any human supervision external inputs. The proposed takes advantage of two types supervisory signals derived from input data: spatiotemporal patterns found between frames single cine (intra-view self-supervision) and interdependencies multiple cines (inter-view self-supervision). combined are learn feature-rich low dimensional embedding space where can be synchronized. Two intra-view self-supervisions used, first based on information encoded by ordering (temporal intra-view) second spatial similarities nearby (spatial intra-view). inter-view self-supervision promote similar embeddings captured same phase views. We evaluate with experiments: 1) Using data 998 patients, Echo-SyncNet shows promising results synchronizing Apical 2 chamber 4 views, which acquired spatially perpendicular each other; 2) 3070 our experiments reveal that learned representations outperform supervised deep method optimized automatic detection fine-grained cycle phase; 3) go one step further show usefulness one-shot scenario key-frame detection. Without fine-tuning, key 1188 validation patient studies identified them only labeled reference cine. do make prior assumption about what specific training, accurately generalize present its training set. Project repository: github.com/fatemehtd/Echo-SyncNet.