Disease classification method and disease classification device

1. A disease classification method, adapted to a disease classification device having a processor, comprising:

inputting, by the processor, samples into a first stage model which is a machine learning model trained by using a data set of basic clinical data and diagnosis results of a plurality of patients for diseases to be classified, to obtain a first stage determination result comprising first samples determined to be positive and fourth samples determined to be negative, wherein a sum of numbers of the first samples and the fourth samples is equal to a number of the samples;

obtaining, by the processor, second samples determined to be positive and third samples determined to be negative via inputting first samples determined to be positive by the first stage model into a second stage specificity model which is a machine learning model trained by using the data set and having a specificity greater than a first threshold and ruling in, by the processor, the second samples;

obtaining, by the processor, fifth samples determined to be positive and sixth samples determined to be negative via inputting the fourth samples determined to be negative by the first stage model into a second stage sensitivity model which is a machine learning model trained by using the data set and having a sensitivity greater than a second threshold and ruling out, by the processor, the sixth samples;

determining, by the processor, a second stage determination result of the second samples and the sixth samples;

inputting, by the processor, the third samples and the fifth samples not ruled in or ruled out into a third stage model which is a machine learning model trained by using the data set and having a LR+, a LR− and a cut point between the LR+, the LR− and the cut point of the second stage specificity model and the second stage sensitivity model to determine a third stage determination result of the third samples and the fifth samples;

determining, by the processor, a first accuracy according to the first stage determination result and actual positive or negative attributes of the samples, and determining, by the processor, a second accuracy according to the second stage determination result and the third stage determination result;

in response to determining that the second accuracy is greater than the first accuracy, providing disease prediction results obtained by applying the second stage specificity model, the second stage sensitivity model and the third stage model; and

in response to determining that the second accuracy is not greater than the first accuracy, discarding the disease prediction results obtained by applying the second stage specificity model, the second stage sensitivity model and the third stage model.

2. The disease classification method according to claim 1 , further comprising:

determining a second output probability of the second stage specificity model or the second stage sensitivity model according to a first output probability of the first stage model and LR+ or LR− of the second stage specificity model or the second stage sensitivity model, and obtaining the second stage determination result according to the second output probability.

3. The disease classification method according to claim 2 , further comprising:

determining a first odds ratio according to a first output probability of a seventh sample among the samples in the first stage model, wherein

a second odds ratio is obtained by multiplying the first odds ratio by the LR+ of the second stage specificity model or the second stage sensitivity model when the seventh sample is determined to be positive by the second stage specificity model or the second stage sensitivity model,

the second odds ratio is obtained by multiplying the first odds ratio by the LR− of the second stage specificity model or the second stage sensitivity model when the seventh sample is determined to be negative by the second stage specificity model or the second stage sensitivity model, and

a second output probability is determined according to the second odds ratio.

4. The disease classification method according to claim 3 , wherein the first odds ratio=the first output probability/(1-the first output probability), and the second output probability=the second odds ratio/(the second odds ratio+1).

5. The disease classification method according to claim 3 , wherein the seventh sample is not ruled out or ruled in by the second stage specificity model or the second stage sensitivity model,

a third odds ratio is obtained by multiplying the second odds ratio by LR+ of the third stage model when the seventh sample is determined to be positive by the third stage model,

the third odds ratio is obtained by multiplying the second odds ratio by LR− of the third stage model when the seventh sample is determined to be negative by the third stage model, and

a third output probability is calculated according to the third odds ratio.

6. The disease classification method according to claim 5 , wherein the third output probability=the third odds ratio/(the third odds ratio+1).

7. A disease classification device, comprising:

a processor; and

a memory coupled to the processor,

wherein the processor is configured to perform:

inputting samples into a first stage model which is a machine learning model trained by using a data set of basic clinical data and diagnosis results of a plurality of patients for diseases to be classified, to obtain a first stage determination result comprising first samples determined to be positive and fourth samples determined to be negative, wherein a sum of numbers of the first samples and the fourth samples is equal to a number of the samples;

obtaining second samples determined to be positive and third samples determined to be negative via inputting first samples determined to be positive by the first stage model into a second stage specificity model which is a machine learning model trained by using the data set and having a specificity greater than a first threshold and ruling in the second samples;

obtaining fifth samples determined to be positive and sixth samples determined to be negative via inputting the fourth samples determined to be negative by the first stage model into a second stage sensitivity model which is a machine learning model trained by using the data set and having a sensitivity greater than a second threshold and ruling out the sixth samples;

determining a second stage determination result of the second samples and the sixth samples;

inputting the third samples and the fifth samples not ruled in or ruled out into a third stage model which is a machine learning model trained by using the data set and having a LR+, a LR− and a cut point between the LR+, the LR− and the cut point of the second stage specificity model to determine a third stage determination result of the third samples and the fifth samples;

determining a first accuracy according to the first stage determination result and actual positive or negative attributes of the samples, and determining a second accuracy according to the second stage determination result and the third stage determination result;

in response to determining that the second accuracy is greater than the first accuracy, providing disease prediction results obtained by applying the second stage specificity model, the second stage sensitivity model and the third stage model; and

in response to determining that the second accuracy is not greater than the first accuracy, discarding the disease prediction results obtained by applying the second stage specificity model, the second stage sensitivity model and the third stage model.