Fig 1.
Overview of the proposed method.
Fig 2.
Left ventricular long-axis view and short-axis papillary muscle level view (left: views of anatomy; right: normal views).
Fig 3.
Selection of the study population.
Table 1.
Baseline clinical characteristics of the study cohort.
Fig 4.
Preprocessing of the input data.
All frames of one cardiac cycle taken from the original echocardiography image were interpolated to 30 frames, and the ECG in the image was removed and trimmed.
Fig 5.
VGG16 structure.
Fig 6.
Simplified diagram of the global average pooling and global max pooling compression methods.
Pooling is performed to extract the maximum or average values from the last feature map of the CNN.
Fig 7.
Schematic diagram of a recurrent neural network.
In the diagram, x is the input, h is the hidden layer, and y is the output. The RNN learns by passing the weights of the hidden layer to the next hidden layer in the forward direction.
Fig 8.
Diagram of the long short-term memory principle.
Input, forgetting, and output gates are used to determine the information to be input into, retained, and output from the cells, enabling the learning of sequential data over a long period of time.
Fig 9.
Classification of myocardial infarction and normal myocardium cases. LSTM, long short-term memory.
Features from 1 to 30 frames were input to each LSTM, and the classification was performed using the softmax function based on the LSTM output.
Fig 10.
Simplified diagram of the cross-validation method (number of folds = 5).
Fig 11.
Schematic representation of the artificial neural network classification of myocardial infarction and normal cases.
Features from 1 to 30 frames were transformed into one-dimensional data and input into an ANN.
Table 2.
Overall classification accuracy of long short-term memory for long-axis view images.
Table 3.
Overall classification accuracy of the artificial neural network for the long-axis view images.
Table 4.
Overall classification accuracy of the long short-term memory for the short-axis papillary muscle level images.
Table 5.
Overall classification accuracy of the artificial neural network for the short-axis papillary muscle level images.
Table 6.
Overall classification accuracy with changing parameters.
Table 7.
Results of long-axis view images.
Table 8.
Results of short-axis papillary muscle view images.
Fig 12.
ROC curves of the LSTM versus ANN in long-axis view.
Fig 13.
ROC curves of the GAP versus GMP in long-axis view.
Fig 14.
ROC curves of the LSTM versus ANN in short-axis papillary muscle view.
Fig 15.
ROC curves of the GAP versus GMP in short-axis papillary muscle view.
Fig 16.
False-positive cases on long-axis view images.
Fig 17.
True-negative cases on long-axis view images.
Fig 18.
False-negative cases on long-axis view images with anteroseptal infarction with regional abnormal wall motion circled in red.
Fig 19.
True-positive cases on long-axis view images.
Fig 20.
False-positive cases on short-axis view papillary muscle level images.
Fig 21.
True-negative cases on short-axis view papillary muscle level images.
Fig 22.
False-negative cases on short-axis view papillary muscle level images.
Fig 23.
True-positive cases on short-axis view papillary muscle level images.
Table 9.
Comparison of related works.