Atrial high rate episode prediction model training method and apparatus

A prediction model training apparatus, comprising:

• a storage, configured to store a plurality of electrocardiograms and a plurality of episode data corresponding to the electrocardiograms, wherein each of the electrocardiograms belongs to an atrial high rate episode state, each of the episode data comprises an episode duration, and the episode duration is configured to indicate a time period of the atrial high rate episode state;
• and
• a processor, electrically connected to the storage, wherein the processor is configured to:
• filtering a plurality of valid electrocardiograms from the electrocardiograms according to an elimination rule, wherein each of the electrocardiograms corresponds to an earliest electrocardiogram segment of each of a plurality of original electrocardiograms, and each of the original electrocardiograms belongs to the atrial high rate episode state;
• generating a plurality of training features corresponding to the valid electrocardiograms according to a training category set, wherein the training category set comprises at least one of a heart rate variability category, a medical record category, an amplitude category, a frequency spectrum category, or a combination thereof;
• and
• training a prediction model according to the valid electrocardiograms, the training features, and the episode data corresponding to the valid electrocardiograms, wherein the prediction model is configured to predict whether the atrial high rate episode state will last more than a time threshold according to an untested electrocardiogram in the atrial high rate episode state;
• wherein the operation of filtering the plurality of valid electrocardiograms from the electrocardiograms according to the elimination rule further comprises:
• eliminating at least one invalid electrocardiogram with heart rates adjusted by a pacemaker from the electrocardiograms;
• eliminating the at least one invalid electrocardiogram without an atrial tachycardia symptom or an atrial fibrillation symptom from the electrocardiograms;
• and
• eliminating the at least one invalid electrocardiogram corresponding to an episode duration lower than the time threshold and a total duration higher than a day threshold from the electrocardiograms;
• wherein if the training category set comprises the frequency spectrum category, the operation of generating the plurality of training features corresponding to the valid electrocardiograms according to the training category set further comprises:
• transforming the valid electrocardiograms into a plurality of energy values on a plurality of specific frequency bands;
• and
• calculating a frequency spectrum distribution of each of the valid electrocardiograms according to the energy values to obtain a plurality of frequency spectrum features corresponding to the valid electrocardiograms, wherein the training features comprise the frequency spectrum features.