MDS malignant transformation risk assessment method based on machine learning and bone marrow cell morphology
By constructing a machine learning-based risk assessment model for the malignant transformation of MDS and using bone marrow smear images to identify cell proportions, the economic and simplicity issues of risk assessment for the transformation of MDS to AML in primary hospitals have been resolved, thereby improving the accuracy of treatment and the survival rate of high-risk patients.
Patent Information
- Application Number
- CN202511649576.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-12
- Publication Date
- 2026-02-13
AI Technical Summary
Existing technologies lack an economical and simple method for assessing the risk of transformation from myelodysplastic syndromes (MDS) to acute myeloid leukemia (AML) in primary care hospitals, resulting in high economic burden and untimely assessment.
By combining machine learning with bone marrow cell morphology data, a risk assessment model for the malignant transformation of MDS was constructed. The model uses bone marrow smear images to identify the proportion of target cells, calculates the risk assessment score, and recommends treatment plans.
This provides a low-cost, readily available risk assessment method that can optimize treatment decisions and improve the survival rate of high-risk patients.
Smart Images
Figure CN121528523A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of bioinformatics, and in particular to a method for evaluating the malignant transformation risk of MDS based on machine learning and bone marrow cell morphology. BACKGROUND
[0002] Myelodysplastic syndrome (MDS) is a heterogeneous clonal hematopoietic disease characterized by ineffective hematopoiesis and dysplasia, and accompanied by the risk of progression to acute myeloid leukemia (AML). About 20%-30% of MDS patients will eventually transform into secondary AML (sAML), which will significantly reduce the survival rate and quality of life of patients. Bone marrow morphology information dominated by bone marrow blast cells plays an important role in the process of MDS transformation into AML. In recent years, the use of machine learning to integrate medical information to establish a prediction model has gradually emerged, while the mechanism of MDS transformation into leukemia is not clear, and the current prediction of the prognosis of such patients is a problem that hematologists are concerned about.
[0003] The IPSS-M prognostic evaluation system proposed in 2022 further introduces gene mutation information in addition to IPSS-R, providing more personalized risk assessment and prognosis for patients, covering clinical endpoints such as survival and AML transformation. However, the large economic burden caused by the high-standard data and high-performance equipment required for the application of IPSS-M limits the accessibility of IPSS-M in some primary hospitals. In terms of bone marrow morphology, 10% is still used as the standard for defining MDS hematopoietic disorders, but when the blast cells are <5%, the National Comprehensive Cancer Network (NCCN) in the United States may be more inclined to include the case of 2%-5% bone marrow blast cells into an earlier or milder disease stage, and WHO classifies it as MDS, which is a divergence between the two, so strict counting of bone marrow blast cells and further differentiation and definition still have certain clinical value for risk stratification of MDS and related diseases. For primary hospitals, using easily available bone marrow morphology information to construct a related prognosis model as a supplement to IPSS-R may be one of the accessible methods at present, but there are still few studies on this aspect. It is urgent to develop more economical and more concise risk prediction models to adapt to the use of primary hospitals. SUMMARY
[0004] The purpose of the present application is to provide a method for evaluating the malignant transformation risk of MDS based on machine learning and bone marrow cell morphology, which uses machine learning technology to combine bone marrow morphology data and clinical information of MDS patients, screens the key features of MDS transformation into AML, and constructs a prediction model to evaluate the risk of MDS patients transforming into AML, so as to benefit related patients in areas with limited medical resources.
[0005] To achieve the above-mentioned purposes, the technical solutions of the present application are as follows: A method for evaluating the malignant transformation risk of MDS based on machine learning and bone marrow cell morphology, the method comprising: S101, constructing a model for evaluating the malignant transformation risk of MDS based on bone marrow morphological cell characteristics; S102, collecting bone marrow smear images of MDS patients, identifying the bone marrow smear images, and obtaining the proportion data of target type bone marrow cells, denoted as bone marrow cell type proportion data; S103, inputting the bone marrow cell type proportion data into the model for evaluating the malignant transformation risk of MDS, calculating and outputting the risk evaluation score of the patient from MDS to acute myeloid leukemia (AML); S104, grouping the patients according to the risk evaluation score and recommending a treatment plan.
[0006] Further, step S101 specifically includes the following operations: S201, collecting sample bone marrow smear images and clinical data of multiple patients; S202, extracting multiple bone marrow morphological cell characteristics of the sample bone marrow smear images; S203, inputting the multiple bone marrow morphological cell characteristics into multiple different feature importance evaluation algorithms to obtain multiple important feature sets; S204, calculating the intersection of the multiple important feature sets, inputting the intersection into a multi-factor Cox regression model, and calculating and obtaining the regression coefficients of each bone marrow morphological cell characteristic in the intersection; S205, constructing a risk evaluation score calculation formula according to the regression coefficients of each bone marrow morphological cell characteristic in the intersection to obtain the model for evaluating the malignant transformation risk of MDS.
[0007] Further, inputting the multiple bone marrow morphological cell characteristics into multiple different feature importance evaluation algorithms to obtain multiple important feature sets specifically includes: S301, inputting the multiple bone marrow morphological cell characteristics into a single-factor Cox regression analysis model to screen out important features related to AML, and outputting a first important feature set; S302, establishing a LASSO regression model, taking the AML transformation time as the dependent variable and the multiple bone marrow morphological cell characteristics as the independent variable, adjusting the regularization strength, selecting the important features corresponding to the optimal lambda value, and outputting a second important feature set; S303, perform feature importance analysis on the plurality of bone marrow morphological cell features using an XGBoost algorithm, select an optimal parameter combination through cross-validation, rank the importance of all bone marrow morphological cell features based on the optimal parameter combination, and select the top 10 features as the third important feature set; S304, input the plurality of bone marrow morphological cell features into a random forest model, sort the feature importance, and select the top 5 bone marrow morphological cell features as the fourth important feature set.
[0008] Further, the intersection of the bone marrow morphological cell features includes the proportion of blast cells, the proportion of immature megakaryocytes, and the proportion of granular megakaryocytes.
[0009] Further, the risk assessment calculation formula is as follows:
[0010] In the above formula, represents the risk assessment score, represents the multi-factor Cox regression coefficient of the th bone marrow morphological cell feature in the intersection, represents the feature value of the th bone marrow morphological cell feature in the intersection.
[0011] Further, the patients are grouped according to the risk assessment score, specifically, the best cutoff value of the risk assessment score is dynamically calculated using the surv_cutpoint function in the R software package survminer, the patients with a risk assessment score greater than the best cutoff value are divided into a high-risk group, and the patients with a risk assessment score lower than the best cutoff value are divided into a low-risk group.
[0012] Further, after constructing the MDS malignant transformation risk assessment model, the performance of the model is evaluated through the training set and the validation set, specifically including the following operations: S401, draw the ROC curve of the risk score and calculate the area under the curve, and record the calculation result as the first AUC value; S402, draw the ROC curve of the risk score and calculate the area under the curve, and record the calculation result as the second AUC; S403, draw the calibration curve, perform Hosmer-Lemeshow test on the calibration curve, and obtain the test result; S404, draw the decision curve, evaluate the clinical net benefit of the MDS malignant transformation risk assessment model under different optimal cutoff values according to the decision curve, and obtain the decision benefit evaluation result; S405, evaluate the performance of the MDS malignant transformation risk assessment model according to the first AUC, the second AUC, the test result, and the decision benefit evaluation result.
[0013] Further, the calibration curve is plotted, which includes the following operations: S501. Sort all data in the training set according to the risk assessment scores of the MDS malignant transformation risk assessment model from low to high. S502. Divide the sorted data into several equal groups; S503. For the data within each group, calculate the average predicted probability of MDS transforming into AML, and the average actual probability of MDS transforming into AML. S504. Using the average predicted probability as the horizontal axis and the average actual probability as the vertical axis, obtain the coordinate points corresponding to each group. Connect the coordinate points corresponding to each group in sequence to obtain the calibration curve.
[0014] Furthermore, step S404 specifically includes the following operations: S601. Set several probability thresholds, wherein the probability thresholds are used to characterize the risk assessment score threshold of patients who need to take treatment measures. S602. For each probability threshold, calculate the net benefit of using the MDS malignant transformation risk assessment model. S603. Calculate the net benefit of adopting a strategy of treating all patients and a strategy of not treating all patients; S604. Plot the decision curve corresponding to the net benefit of using the MDS malignant transformation risk assessment model with the probability threshold as the horizontal axis and the net benefit as the vertical axis, and draw two reference curves corresponding to the net benefits of adopting the strategy of treating all patients and the strategy of not treating all patients respectively. S605. Identify the positional relationship between the decision curve and the reference curve, and generate and output the decision benefit evaluation results based on the identification results.
[0015] Compared with the prior art, the beneficial effects of the present invention are: This invention provides a method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology. It establishes a risk assessment model for MDS malignant transformation based on bone marrow morphological cell characteristics. When assessing the risk of a patient's MDS transforming into AML, it only requires collecting the patient's bone marrow smear image and identifying the proportion of target type bone marrow cells to calculate the patient's risk assessment score for MDS transformation into AML. This method has advantages such as easy access at the grassroots level, low economic cost, and ease of calculation. By grouping patients according to the risk assessment score and recommending treatment plans, treatment decisions can be optimized, and the survival rate of high-risk patients can be improved. Attached Figure Description
[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0017] Figure 1 This is a schematic diagram of the overall process of a method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology, provided in an embodiment of the present invention.
[0018] Figure 2 This is a schematic diagram of the construction process of the MDS malignant transformation risk assessment model provided in the embodiment of the present invention.
[0019] Figure 3 This is a schematic diagram of the important feature set filtering process provided in the embodiments of the present invention.
[0020] Figure 4 This is a schematic diagram of the model performance evaluation process provided in the embodiments of the present invention.
[0021] Figure 5 This is a schematic diagram of the calibration curve plotting process provided in an embodiment of the present invention.
[0022] Figure 6 This is a schematic diagram of the decision curve drawing process provided in an embodiment of the present invention. Detailed Implementation
[0023] The principles and features of the present invention are described below with reference to the accompanying drawings. The listed embodiments are only used to explain the present invention and are not intended to limit the scope of the present invention.
[0024] Reference Figure 1 This embodiment provides a method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology. The method includes the following steps: S101. Based on the morphological cell characteristics of bone marrow, a risk assessment model for malignant transformation of myelodysplastic syndrome (MDS) was constructed.
[0025] S102. Collect bone marrow smear images of MDS patients, identify the bone marrow smear images, and obtain the proportion data of target type bone marrow cells, which are recorded as bone marrow cell type proportion data.
[0026] S103. Input the bone marrow cell type ratio data into the MDS malignant transformation risk assessment model, calculate and output the risk assessment score of the patient's transformation from MDS to acute myeloid leukemia (AML).
[0027] S104. Group patients according to their risk assessment scores and recommend treatment plans.
[0028] As one possible implementation method, refer to Figure 2 Step S101 specifically includes the following operations: S201. Collect bone marrow smear images and clinical data from multiple patients.
[0029] For example, the clinical data refers to the patient's general clinical information, including but not limited to: age, height and weight, peripheral blood cell count (white blood cells, monocytes, neutrophils, platelets), hemoglobin, reticulocyte ratio, lactate dehydrogenase, coexisting second tumor state, ferritin, serum iron, etc.; as well as disease-related information, such as FAB classification, triglyceride level, IPSS score, IPSS-R score, and bone marrow karyotype.
[0030] In this implementation, the collected data is used to divide the data into a training set and a validation set to facilitate subsequent model training and validation. For example, 75% of the data is allocated to the training set, and the remaining data is allocated to the validation set.
[0031] S202. Extract multiple bone marrow morphological cell features from bone marrow smear images.
[0032] In this embodiment, the bone marrow morphological cell characteristics include, but are not limited to: the proportion of primitive granulocytes, the proportion of promyelocytes, the proportion of intermediate neutrophils, the proportion of late neutrophils, the proportion of band neutrophils, the proportion of lobulated neutrophils, the proportion of intermediate eosinophils, the proportion of late eosinophils, the proportion of band eosinophils, the proportion of lobulated eosinophils, the proportion of intermediate eosinophils, the proportion of late eosinophils, the proportion of intermediate eosinophils, the proportion of band eosinophils, the proportion of lobulated eosinophils, the proportion of primitive erythrocytes, the proportion of early erythroblasts, the proportion of intermediate erythroblasts, the proportion of late erythroblasts, the granulocyte-erythrocyte ratio, the proportion of prolymphocytes, the proportion of immature lymphocytes, the proportion of mature lymphocytes, the proportion of primitive monocytes, the proportion of immature monocytes, the proportion of mature monocytes, the proportion of primitive plasma cells, the proportion of immature plasma cells, the proportion of mature plasma cells, the number of immature megakaryocytes, the number of granular megakaryocytes, the number of platelet-producing megakaryocytes, and the number of naked megakaryocytes.
[0033] In this embodiment, the extraction of bone marrow morphological cell features can be performed manually, for example, by having two or more experienced laboratory physicians independently review the bone marrow smears of each patient and using the average of the two physicians' reviews as the final bone marrow morphological cell features for subsequent analysis; alternatively, it can be achieved by training a neural network model to automatically extract bone marrow morphological cell features from sample bone marrow smear images. This embodiment does not impose any specific limitations on this method.
[0034] S203. Input multiple bone marrow morphological cell features into multiple different feature importance evaluation algorithms to obtain multiple sets of important features.
[0035] S204. Calculate the intersection of multiple important feature sets, input the intersection into a multifactor Cox regression model, and calculate and obtain the regression coefficients of each bone marrow morphological cell feature in the intersection.
[0036] S205. Based on the regression coefficients of various bone marrow morphological cell characteristics in the intersection, construct the risk assessment score calculation formula to obtain the MDS malignant transformation risk assessment model.
[0037] In step S203, multiple bone marrow morphological cell features are input into multiple different feature importance evaluation algorithms to obtain multiple sets of important features, referring to... Figure 3 Specifically, this includes the following operations: S301. Input multiple bone marrow morphological cell characteristics into a univariate Cox regression analysis model, screen out important features related to AML, and output the first set of important features.
[0038] S302. Establish a LASSO (Least Absolute Shrinkage and Selection Operator) regression model, with AML conversion time as the dependent variable and multiple bone marrow morphological cell characteristics as independent variables. By adjusting the regularization strength, select the important features corresponding to the optimal λ value and output them as the second important feature set.
[0039] S303. Use the XGBoost algorithm to perform feature importance analysis on multiple bone marrow morphological cell features, select the optimal parameter combination through cross-validation, rank all bone marrow morphological cell features based on the optimal parameter combination, and select the top 10 features as the third most important feature set.
[0040] S304. Input multiple bone marrow morphological cell features into a random forest model, rank the features by importance, and select the top 10 bone marrow morphological cell features as the fourth important feature set.
[0041] By calculating the intersection of the first, second, third, and fourth important feature sets, the bone marrow morphological cell characteristics in the intersection are ultimately determined to include the proportions of primitive cells, immature megakaryocytes, and granular megakaryocytes. In other words, in step S102, the proportion data of target type bone marrow cells obtained by identifying the bone marrow smear image should at least include the proportions of primitive cells, immature megakaryocytes, and granular megakaryocytes. These proportion data are calculated using the following formula:
[0042] In the above formula, This data represents the proportion of target type bone marrow cells. Indicates the total number of target type bone marrow cells. Indicates the number of immature megakaryocytes. This indicates the number of granular megakaryocytes. Indicates the number of platelet-producing megakaryocytes. This indicates the number of naked megakaryocytes.
[0043] As another possible implementation, after inputting the bone marrow cell type ratio data into the MDS malignant transformation risk assessment model, the risk assessment score of the patient's transformation from MDS to AML is calculated, specifically using the following formula: .
[0044] In the above formula, Indicates the risk assessment score. The multivariate Cox regression coefficient represents the i-th bone marrow morphological cell characteristic in the intersection set. This represents the feature value of the i-th bone marrow morphological cell feature in the intersection set.
[0045] As another possible implementation, the grouping of patients based on risk assessment scores specifically involves using the surv_cutpoint function in the R package survminer to dynamically calculate the optimal cutoff value for the risk assessment score, classifying patients with risk assessment scores greater than the optimal cutoff value into the high-risk group, and classifying patients with risk assessment scores lower than the optimal cutoff value into the low-risk group.
[0046] In this implementation, the optimal cutoff value is not constant but changes with the data distribution. By periodically adjusting and evaluating the optimal cutoff value, the accuracy and stability of the model can be maintained, thereby more accurately assessing the patient's condition and providing more reasonable treatment measures for the patient.
[0047] As another possible implementation method, refer to Figure 4 After constructing the MDS malignant transformation risk assessment model, the model performance is evaluated using the training and validation sets, specifically including the following operations: S401. Plot the ROC curve of the hazard score and calculate its area under the curve. Record the calculation result as the first AUC value.
[0048] S402. Plot the ROC curve for the hazard stratification and calculate its area under the curve. Record the calculation result as the second AUC.
[0049] S403. Plot the calibration curve and perform the Hosmer-Lemeshow test on the calibration curve to obtain the test results.
[0050] S404. Plot the decision curve, evaluate the clinical net benefit of the MDS malignant transformation risk assessment model at different optimal cutoff values based on the decision curve, and obtain the decision benefit assessment results.
[0051] S405. Based on the first AUC, the second AUC, the test results, and the decision benefit evaluation results, evaluate the performance of the MDS malignant transformation risk assessment model.
[0052] In this implementation, by combining the first AUC, the second AUC, the test results, and the decision benefit assessment results, the model performance can be reflected from different dimensions: a high first AUC indicates that the model has strong core discrimination ability, while a low first AUC indicates that the model's discrimination ability is insufficient and needs further improvement. A high second AUC indicates that the model can effectively group patients by risk, with direct clinical improvement significance. Good test results indicate that the model's assessment results are highly reliable, while poor test results indicate that the model needs recalibration. The decision curve reflects whether the model has clear clinical value. If the decision curve shows a significant net benefit across a broad threshold range, it indicates that making decisions based on the model's assessment results—for example, taking further intervention measures for high-risk patients—is more beneficial than harmful.
[0053] In step S401, plotting the ROC curve of the hazard score and calculating its area under the curve can be achieved using statistical software, such as the roc_curve and auc functions in the sklearn library in Python, which can conveniently calculate the ROC curve and its area under the curve.
[0054] Reference Figure 5 In step S403, the calibration curve is plotted, which specifically includes the following operations: S501. Sort all data in the training set according to the risk assessment scores of the MDS malignant transformation risk assessment model from low to high.
[0055] S502. Divide the sorted data into several equal groups.
[0056] S503. For the data within each group, calculate the average predicted probability of MDS transforming into AML, and the average actual probability of MDS transforming into AML.
[0057] S504. Using the average predicted probability as the horizontal axis and the average actual probability as the vertical axis, obtain the coordinate points corresponding to each group. Connect the coordinate points corresponding to each group in sequence to obtain the calibration curve.
[0058] The calibration curve assesses the model's calibration accuracy by comparing the predicted probabilities with the observed actual probabilities. A perfectly calibrated model should have a calibration curve that runs diagonally from the origin to (1,1). Therefore, after plotting the calibration curve, a diagonal line can be drawn for comparison to determine the model's calibration status.
[0059] Reference Figure 6 In step S404, the clinical net benefit of the evaluation model under different risk probability thresholds is assessed by plotting decision curves, specifically including the following operations: S601. Set several probability thresholds, which are used to characterize the risk assessment score threshold of patients requiring treatment. That is, when a patient's risk assessment score is greater than the threshold, treatment is administered to the patient.
[0060] S602. For each probability threshold, calculate the net benefit using the MDS malignant transformation risk assessment model.
[0061] For example, the net benefit of using the MDS malignant transformation risk assessment model is calculated as follows:
[0062] In the above formula, This indicates the net benefit calculated using the MDS malignant transformation risk assessment model. This represents the number of patients who have actually converted and are recommended for treatment. This indicates the number of patients who did not convert but were advised to receive treatment. Indicates the total number of patients. This represents the probability threshold.
[0063] S603. Calculate the net benefit of adopting a strategy of treating all patients and a strategy of not treating all patients.
[0064] In this implementation, the net benefit of treating all patients is calculated in the same way as the net benefit using the MDS malignant transformation risk assessment model, the difference being that TP represents the actual number of transformed patients, while FP represents the number of non-transformed patients. The net benefit of not treating all patients is 0.
[0065] S604. Plot the decision curve corresponding to the net benefit using the MDS malignant transformation risk assessment model with the probability threshold as the horizontal axis and the net benefit as the vertical axis, and draw two reference curves corresponding to the net benefits of the strategy of treating all patients and the strategy of not treating all patients.
[0066] S605. Identify the positional relationship between the decision curve and the reference curve, and generate and output the decision benefit evaluation results based on the identification results.
[0067] In this implementation, by observing the positional relationship between the decision curve and the reference curve, it is possible to intuitively determine the range of thresholds within which the model has clinical applicability, and the extent of its applicability.
[0068] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology, characterized in that, The method includes: S101. Based on bone marrow morphological cell characteristics, a risk assessment model for malignant transformation of myelodysplastic syndrome (MDS) was constructed. S102. Collect bone marrow smear images of MDS patients, identify the bone marrow smear images, obtain the proportion data of target type bone marrow cells, and record them as bone marrow cell type proportion data. S103. Input the bone marrow cell type ratio data into the MDS malignant transformation risk assessment model, calculate and output the risk assessment score of the patient's transformation from MDS to acute myeloid leukemia (AML). S104. Group patients according to their risk assessment scores and recommend treatment plans.
2. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 1, characterized in that, Step S101 specifically includes the following operations: S201. Collect bone marrow smear images and clinical data from multiple patients; S202. Extract multiple bone marrow morphological cell features from bone marrow smear images; S203. Input multiple bone marrow morphological cell features into multiple different feature importance evaluation algorithms to obtain multiple sets of important features; S204. Calculate the intersection of multiple important feature sets, input the intersection into a multifactor Cox regression model, and calculate and obtain the regression coefficients of each bone marrow morphological cell feature in the intersection. S205. Based on the regression coefficients of various bone marrow morphological cell characteristics in the intersection, construct the risk assessment score calculation formula to obtain the MDS malignant transformation risk assessment model.
3. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 2, characterized in that, Multiple bone marrow morphological cell features are input into several different feature importance evaluation algorithms to obtain multiple sets of important features, specifically including: S301. Input multiple bone marrow morphological cell features into a univariate Cox regression analysis model, screen out important features related to AML, and output the first set of important features. S302. Establish a LASSO regression model with AML conversion time as the dependent variable and multiple bone marrow morphological cell characteristics as independent variables. Select the optimal model by adjusting the regularization strength. The important features corresponding to the values are output as the second most important feature set; S303. Use the XGBoost algorithm to perform feature importance analysis on multiple bone marrow morphological cell features, select the optimal parameter combination through cross-validation, rank all bone marrow morphological cell features based on the optimal parameter combination, and select the top 10 features as the third most important feature set. S304. Input multiple bone marrow morphological cell features into a random forest model, rank the features by importance, and select the top 5 bone marrow morphological cell features as the fourth important feature set.
4. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 2, characterized in that, The bone marrow morphological cell characteristics in the intersection include the proportion of primitive cells, immature megakaryocytes, and granular megakaryocytes.
5. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 2, characterized in that, The risk assessment calculation formula is as follows: In the above formula, Indicates the risk assessment score. The multivariate Cox regression coefficient represents the i-th bone marrow morphological cell characteristic in the intersection set. This represents the feature value of the i-th bone marrow morphological cell feature in the intersection set.
6. A method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 1 or 5, characterized in that, The grouping of patients based on risk assessment scores involves using the surv_cutpoint function in the R package survminer to dynamically calculate the optimal cutoff value for the risk assessment score. Patients with risk assessment scores greater than the optimal cutoff value are assigned to the high-risk group, while patients with risk assessment scores lower than the optimal cutoff value are assigned to the low-risk group.
7. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 6, characterized in that, After constructing the MDS malignant transformation risk assessment model, the model performance is evaluated using the training and validation sets, specifically including the following operations: S401. Plot the ROC curve of the hazard score and calculate its area under the curve. Record the calculation result as the first AUC value. S402. Plot the ROC curve for hazard stratification and calculate its area under the curve. Record the calculation result as the second AUC. S403. Plot the calibration curve, perform the Hosmer-Lemeshow test on the calibration curve, and obtain the test results; S404. Plot the decision curve, evaluate the clinical net benefit of the MDS malignant transformation risk assessment model at different optimal cutoff values based on the decision curve, and obtain the decision benefit assessment results. S405. Based on the first AUC, the second AUC, the test results, and the decision benefit evaluation results, evaluate the performance of the MDS malignant transformation risk assessment model.
8. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 7, characterized in that, Plotting the calibration curve involves the following steps: S501. Sort all data in the training set according to the risk assessment scores of the MDS malignant transformation risk assessment model from low to high. S502. Divide the sorted data into several equal groups; S503. For the data within each group, calculate the average predicted probability of MDS transforming into AML, and the average actual probability of MDS transforming into AML. S504. Using the average predicted probability as the horizontal axis and the average actual probability as the vertical axis, obtain the coordinate points corresponding to each group. Connect the coordinate points corresponding to each group in sequence to obtain the calibration curve.
9. The method for assessing the risk of malignant transformation of MDS based on machine learning and bone marrow cell morphology according to claim 7, characterized in that, Step S404 specifically includes the following operations: S601. Set several probability thresholds, wherein the probability thresholds are used to characterize the risk assessment score threshold of patients who need to take treatment measures. S602. For each probability threshold, calculate the net benefit of using the MDS malignant transformation risk assessment model. S603. Calculate the net benefit of adopting a strategy of treating all patients and a strategy of not treating all patients; S604. With the probability threshold as the horizontal axis and the net benefit as the vertical axis, plot the decision curve corresponding to the net benefit of using the MDS malignant transformation risk assessment model, and two reference curves corresponding to the net benefits of adopting the strategy of treating all patients and the strategy of not treating all patients. S605. Identify the positional relationship between the decision curve and the reference curve, and generate and output the decision benefit evaluation results based on the identification results.