A novel kernel-based classification algorithm using low-level image features can accurately and efficiently classify 2D echocardiogram views without requiring pixel-level segmentation.
This paper presents a view classification method for 2D heart ultrasound. Our method uses low-level image features to train a frame-level classifier, which unlike related approaches, does not require an additional pixel-level classification of heart structures. By employing kernel-based classification, our algorithm can classify images from any phase of the heartbeat cycle and efficiently incorporate information from subsequent frames without re-training the model. On real-world data, our algorithm achieves 98.51% accuracy for 8-way classification. While the method can efficiently aggregate multiple frames, in ~94% of the tests, identification only requires a single frame.
Wu et al. (Mon,) studied this question.