System and method for pulmonary embolism detection from the electrocardiogram using deep learning
A method of assessing a likelihood of a patient having a pulmonary embolism (PE) comprises receiving discrete patient data, including patient-related clinical and demographic data pertinent to the patient, receiving electrocardiograph (ECG) waveform data obtained from examination of the patient, processing both the received discrete patient data and received set of ECG waveform data using a supervised deep learning multimodal fusion model that has been trained using analogous input training data including both discrete patient data and ECG waveform data to obtain an optimal match to results from corresponding patient computed tomography pulmonary angiograms (CTPA), indicative of a presence or absence of a pulmonary embolism, and outputting a measure of the likelihood of the patient having a pulmonary embolism.
1. A method for generating and providing an output associated with a determined likelihood of a patient having a pulmonary embolism, the method comprising:
performing, by one or more computing devices, operations comprising:
accessing discrete patient data, including clinical and demographic information associated with the patient;
accessing electrocardiograph (ECG) waveform data obtained from examination of the patient;
computing first ECG waveform features by applying a deep learning neural network (DNN) model to the accessed ECG waveform data;
deriving second ECG waveform features by reducing a dimensionality of the first ECG waveform features;
providing the second ECG waveform features and at least some of the accessed discrete patient data as input to a multimodal fusion model;
receiving in response, from the multimodal fusion model, fusion model output representing a likelihood of the patient having a pulmonary embolism;
generating, based on the likelihood of the patient having the pulmonary embolism, an output representing a recommendation whether to order a computed tomography pulmonary angiography (CTPA) scan; and
providing the output representing the recommendation.
2. The method of claim 1 , wherein the DNN model and the multimodal fusion model have been trained via semi-supervised deep learning.
3. The method of claim 1 , wherein the DNN model and the multimodal fusion model have been trained via self-supervised deep learning.
4. The method of claim 1 , wherein reducing the dimensionally of the output of the DNN model comprises performing principal component analysis (PCA).
5. The method of claim 4 , wherein the PCA reduces the dimensionality of the output of the DNN model to a selected number of ECG waveform features.
6. The method of claim 1 , wherein the DNN comprises a convolutional neural network.
7. The method of claim 1 , wherein the DNN comprises at least one of a recurrent neural network, a long-term-short-term memory network, a graph neural network, or a transformer network.
8. The method of claim 1 , the operations further comprising:
determining that a measure of risk of the patient having a pulmonary embolism exceeds a preset threshold,
wherein, in response to the measure of risk being at or above the preset threshold, the generated output represents a positive recommendation to order a CTPA scan.
9. The method of claim 1 , the operations further comprising:
determining that a measure of risk of the patient having a pulmonary embolism exceeds a preset threshold; and
arranging for medical treatment in response to the measure of risk being at or above the preset threshold.
10. The method of claim 1 , wherein the accessed discrete patient data includes at least a brain natriuretic peptide (BNP) level, a D-dimer level, and a troponin level.
11. The method of claim 1 , wherein the accessed discrete patient data includes at least a brain natriuretic peptide (BNP) level, D-dimer level, troponin level, age, sex, and comorbidity information.
12. The method of claim 1 , wherein, in the training data, for any patient encounter where the CTPA data was obtained with a defined period of the ECG waveform data and is positive for pulmonary embolism, the ECG waveform data of the input data is labeled as PE positive.
13. The method of claim 1 , wherein the accessed ECG waveform data includes an ECG obtained over a selected range of data sampling frequencies.
14. The method of claim 1 , wherein the accessed ECG waveform data includes multiple ECG waveforms taken from the patient.