Atrial fibrillation evaluation model training method, and atrial fibrillation evaluation method and device

Abstract

The invention discloses an atrial fibrillation evaluation model training method, and an atrial fibrillation evaluation method and device, and belongs to the technical field of intelligent medical treatment. The atrial fibrillation evaluation method comprises the steps that a face image including a training object is obtained, a VPPG pulse signal of the face of the training object is extracted through a video photoplethysmography method, the pulse signal is input into a noise reduction self-encoder based on a recurrent neural network, and pulse signal rhythm change features are extracted; and the pulse features are input into an atrial fibrillation classification model under the constraint of a feature scoring criterion to realize detection of the atrial fibrillation. According to the atrial fibrillation evaluation method, distorted face pulse signals can be coded into the non-interference pulse feature, and the pulse features in an atrial fibrillation attack period are emphatically monitored, so that the robustness and the accuracy of paroxysmal atrial fibrillation detection are improved, and a convenient way is provided for rapid screening of atrial fibrillation in a real environment.

Description

technical field [0001] The invention belongs to the technical field of intelligent medical treatment, and in particular relates to an atrial fibrillation evaluation model training method, an atrial fibrillation evaluation method and a device. Background technique [0002] Atrial fibrillation, abbreviated as atrial fibrillation, is a common arrhythmia disease in which the heart loses its normal systolic function and vibrates disorderly. Stroke is a complication of atrial fibrillation. Patients with atrial fibrillation are five times more likely to suffer a stroke than the general population, with approximately 20% of strokes directly caused by atrial fibrillation. Therefore, timely detection and treatment of AF is a key step in stroke prevention. However, early atrial fibrillation is often paroxysmal atrial fibrillation without persistent symptoms. Ordinary short-term ECG examination may not be able to capture the onset moment of atrial fibrillation, thus affecting the det...

AI Technical Summary

Problems solved by technology

[0005] Aiming at the problem that the pulse signal existing in the prior art has different degrees of distortion, resulting in low detection accuracy of atrial fibrillation, the present invention provides a method for training an atrial fibrillation evaluation model, first extracting the VPPG pulse signal of the face of the training object; Further extract the rhythm change feature of the VPPG pulse signal as training data, input the softmax classifier to train and obtain the atrial fibrillation evaluation model, the PPG pulse signal reconstructed by the rhythm change feature does not contain distortion components, and the movement artifacts in the VPPG pulse signal The shadow has anti-interference ability; and the characteristic dimension of the rhythm change is lower than that of the VPPG pulse signal, which is a compression of the redundancy of the VPPG pulse signal. Compared with the VPPG pulse signal, it has better expressive ability and robustness, thus more Applicable to the identification of atrial fibrillation

Images