WO2024053925A1

WO2024053925A1 - Method and device for predicting amyloid pet positivity in patients suspected of amyloid angiopathy

Info

Publication number: WO2024053925A1
Application number: PCT/KR2023/012891
Authority: WO
Inventors: 정영희
Original assignee: 의료법인 명지의료재단
Priority date: 2022-09-05
Filing date: 2023-08-30
Publication date: 2024-03-14
Also published as: KR102517529B1

Abstract

The present invention relates to a method for predicting amyloid PET positivity in patients suspected of amyloid angiopathy, using an amyloid PET positivity prediction device, the method comprising the steps of: receiving brain imaging data for the patient in the amyloid PET positivity prediction device; performing machine learning on the brain imaging data received in the amyloid PET positivity prediction device; and predicting the amyloid PET positivity rate for the patient through machine learning in the amyloid PET positivity prediction device. According to the present invention, amyloid PET positivity is predicted using brain imaging for patients suspected of amyloid angiopathy, whereby it is possible to reduce the costs of amyloid PET scanning and prevent issues related to radiation exposure.

Description

Method and device for predicting amyloid PET positivity for patients suspected of amyloid angiopathy

The present invention relates to a method and device for predicting amyloid PET positivity, and more specifically, to a method and device for predicting amyloid PET positivity for patients suspected of amyloid angiopathy.

With recent advances in scientific technology and imaging techniques, brain imaging techniques for diagnosing and predicting brain diseases include computed tomography (CT), magnetic resonance imaging (MRI), and positron tomography imaging. emission tomography (PET), etc. have been developed. Among these, the amyloid PET test, which can check the accumulation of amyloid beta (β), is known to be effective in diagnosing degenerative brain diseases such as Alzheimer's disease and dementia.

Alzheimer's disease is a representative degenerative brain disease that accounts for about 70% of dementia, and its core pathology is the accumulation of beta-amyloid protein in the cerebrum. For an accurate diagnosis of Alzheimer's disease, it is very important to identify cerebral beta-amyloid accumulation. However, for a long time, cerebral beta-amyloid accumulation lesions could only be confirmed through brain autopsy after the patient died, and it was difficult to confirm in a living state, making it difficult to confirm Alzheimer's disease. However, in 2004, William E. Klunk, a geriatric psychiatrist at the University of Pittsburgh, and Chester A. Mathis, a radiochemist, developed a substance called Pittsburgh Compound B (PiB), a positron emission tomography (PET) ligand. . PiB PET imaging using this technology made it possible to accurately determine the presence and amount of beta-amyloid accumulation in the brain of a living person, marking a groundbreaking turning point in Alzheimer's disease treatment and research. Afterwards, in addition to PiB, additional ligands that can identify beta amyloid, such as florbetaben, were developed and reported to have similar accuracy. In fact, amyloid PET has been approved for clinical use by the U.S. FDA and Korea's KFDA to check for beta-amyloid accumulation in patients with cognitive impairment.

The amyloid PET test is performed by injecting a special contrast agent that reacts only to amyloid into the body to determine whether an abnormal form of amyloid, which is produced due to abnormalities in the protein metabolism process, is deposited within the brain. Then, a PET scan is taken to determine whether the deposit is present. It determines whether it is positive or negative.

Amyloid beta accumulation is known to affect cognitive function, but early detection is difficult because these differences appear subtly in preclinical Alzheimer's disease. Meanwhile, it is difficult to perform many amyloid PET tests due to concerns about high cost and radioactive tracer problems. In other words, the amyloid PET (PET) test cost is high, there is a problem with radiation exposure, and in particular, there are many cases where the amyloid PET is rejected during clinical trials because it is negative. Therefore, there is a need for a way for clinicians to identify Alzheimer's disease at an early stage in preclinical stages.

The present invention was developed to solve the above problems, and its purpose is to provide a positive amyloid PET prediction model for patients suspected of amyloid angiopathies using machine learning.

The object of the present invention is not limited to the object mentioned above, and other objects not mentioned will be clearly understood by those skilled in the art from the description below.

The present invention to achieve this purpose relates to a method for predicting amyloid PET positivity for a patient suspected of amyloid angiopathy in an amyloid PET positivity prediction device, wherein the amyloid PET positivity prediction device receives brain imaging data for the patient. A step of performing machine learning on brain image data received from the amyloid PET positivity prediction device, and predicting an amyloid PET positivity rate for the patient through the machine learning in the amyloid PET positivity prediction device.

The brain imaging data may be brain MRI (magnetic resonance imaging) data.

The machine learning may be a gradient boosting machine (GBM) algorithm or a random forest (RF) algorithm.

In order to predict the amyloid PET positivity rate, machine learning can be performed using the number of lobar cerebral microbleeds, lacunes, patient's age, and microbleeds in the dentate nucleus as variables. .

In the amyloid PET positive prediction device for a patient suspected of amyloid angiopathy of the present invention, a receiver for receiving brain image data for the patient, performing machine learning on the brain image data received from the receiver, and performing the machine learning. It includes a prediction unit for predicting the amyloid PET positivity rate for the patient and a display unit for displaying the amyloid PET positivity rate predicted by the prediction unit.

The brain imaging data may be brain MRI (magnetic resonance imaging) data.

In order to predict the amyloid PET positivity rate, the prediction unit performs machine learning using the number of lobar cerebral microbleeds, lacunes, patient's age, and microbleeds in the dentate nucleus as variables. can do.

According to the present invention, by predicting amyloid PET positivity using brain imaging for patients suspected of amyloid angiopathies, the cost of amyloid PET testing can be reduced and radiation exposure problems can be prevented. In addition, according to the present invention, the dropout rate due to negative amyloid PET during clinical testing can be reduced.

In addition, according to the present invention, by predicting amyloid PET positivity, there is an effect of helping to predict the patient's cognitive function or prognosis.

Figure 1 is a block diagram showing the internal configuration of an amyloid pet positive prediction device according to an embodiment of the present invention.

Figure 2 is a flowchart showing a method for predicting amyloid PET positivity in the amyloid PET positivity prediction device according to an embodiment of the present invention.

Figure 3 is a flowchart showing the amyloid PET positive prediction process according to an embodiment of the present invention.

Figure 4 is a chart summarizing the clinical characteristics of patients who participated in the experiment of the present invention.

Figure 5 is a graph showing the importance of variables in the GBM model and RF model in the experiment of the present invention.

Figure 6 is a graph showing optimal threshold values detected at multiple change points in the GBM model and RF model in the experiment of the present invention.

Figure 7 is a table showing cut-off values in the GBM model and RF model in the experiment of the present invention.

Figure 8 is a graph showing cutoff values of important variables in the GBM model and RF model in the experiment of the present invention.

Figure 9 is a chart summarizing the performance of the GBM model and the RF model in the experiment of the present invention.

Referring to FIG. 1, in the present invention, the apparatus 100 for predicting positive amyloid PET for a patient suspected of amyloid angiopathy includes a receiving unit 110, a prediction unit 120, and a display unit 130.

The receiving unit 110 serves to receive brain image data 10 about the patient. In one embodiment of the present invention, the brain imaging data 10 may be brain MRI (magnetic resonance imaging) data.

The prediction unit 120 performs machine learning on the brain image data received from the receiver 110 and predicts the amyloid PET positivity rate for the patient through machine learning.

The display unit 130 serves to display the amyloid PET positivity rate predicted by the prediction unit 120.

In one embodiment of the present invention, machine learning may be a gradient boosting machine (GBM) algorithm or a random forest (RF) algorithm.

In order to predict the amyloid PET positivity rate, the prediction unit 120 performs machine learning using the number of lobar cerebral microbleeds, lacunes, patient's age, and microbleeds in the dentate nucleus as variables. can be performed.

Referring to FIG. 2, the method for predicting amyloid PET positivity includes receiving brain imaging data 10 for a patient from the amyloid PET positivity prediction device 100 (S201), and receiving brain image data 10 for the patient from the amyloid PET positivity prediction device 100. It includes performing machine learning on brain image data (S203) and predicting the amyloid PET positivity rate for the patient through machine learning in the amyloid PET positivity prediction device 100 (S205).

In one embodiment of the present invention, brain imaging data may be brain MRI (magnetic resonance imaging) data.

In order to predict the amyloid PET positivity rate in the present invention, machine learning is performed using the number of lobar cerebral microbleeds, lacunes, patient's age, and microbleeds in the dentate nucleus as variables. You can.

Referring to FIG. 3, when brain image data is received from the receiver 110 (S301), the prediction unit 120 analyzes the brain image data and determines that the number of lobar microbleeds is 14 or more (S303) and the number of lacunar cerebral infarcts is 7. If the number is less than 100% (S305), the patient is over 64 years old (S307), and there is no microhemorrhage in the dentate nucleus (S309), it is judged to be amyloid PET positive (S311).

Since the present invention can make various changes and have various embodiments, specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the present invention to specific embodiments, and should be understood to include all changes, equivalents, and substitutes included in the spirit and technical scope of the present invention.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise. In this application, terms such as “comprise” or “have” are intended to designate the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, but are not intended to indicate the presence of one or more other features. It should be understood that this does not exclude in advance the possibility of the existence or addition of elements, numbers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by a person of ordinary skill in the technical field to which the present invention pertains. Terms such as those defined in commonly used dictionaries should be interpreted as having meanings consistent with the meanings they have in the context of the related technology, and should not be interpreted as having ideal or excessively formal meanings, unless explicitly defined in the present application. No.

In addition, when describing with reference to the accompanying drawings, identical components will be assigned the same reference numerals regardless of the reference numerals, and overlapping descriptions thereof will be omitted. In describing the present invention, if it is determined that a detailed description of related known technologies may unnecessarily obscure the gist of the present invention, the detailed description will be omitted.

The present invention relates to a method and device for predicting amyloid PET positivity for patients suspected of amyloid angiopathy.

We will now describe the experimental process and results according to an embodiment of the present invention.

In the present invention, two interpretable machine learning algorithms, GBM (Gradient Boosting Machine) and GBM (Gradient Boosting Machine), are used to predict amyloid-β (Aβ) PET positivity in patients with suspected cerebral amyloid angiopathy (CAA) MRI markers. RF (Random Forest) was applied. The most influential variables in predicting Aβ positivity in the GBM algorithm and RF algorithm were the number of lobar cerebral microbleeds (CMBs), lacunes, CMBs in the dentate nucleus, and patient age ( age) was selected.

In the present invention, using machine learning that can be interpreted in the environment, clinical usefulness was demonstrated by quantifying the relative importance and cutoff value of predictive variables for Aβ positivity in patients suspected of having CAA markers.

Cerebral amyloid angiopathy (CAA) is a cerebral small vessel disease (CSVD) characterized by amyloid β (Aβ) deposition in pial and cortical blood vessels1,2.

Recently, the clinical utility of Aβ proton emission tomography (PET) in patients with CAA has been widely investigated. Aβ-positive (+) PET scans in patients with CAA MRI markers may have clinical utility in two aspects. First, Aβ positivity in CAA patients allows clinicians to predict the prognosis of cognitive trajectory. Previously, it has been shown that Aβ+ patients with probable CAA have a worse cognitive trajectory than their Aβ negative (-) counterparts. Second, Aβ PET positivity may provide insight into the underlying vascular pathology in patients with suspected CAA MRI markers. there is. Therefore, predicting Aβ positivity in patients with CAA MRI markers would be clinically useful because it may help predict prognosis.

Among prediction models, machine learning methods are receiving a lot of attention due to their high predictive power and stable performance. However, the lack of interpretability of internal processing has become a major problem in machine learning research. To overcome this limitation, two tree-based machine learning models, GBM (Gradient Boosting Machine)15 and Random Forest (RF), were selected in the present invention. These two methods can effectively quantify the relative importance of variables and provide cutoff values that provide clinically meaningful insights.

Therefore, in the present invention, we use a machine learning-based model to identify the most important variables and their optimal cutoff values (e.g., number of lobar microbleeds) to predict Aβ PET positivity.

Figure 4 is a chart summarizing the clinical characteristics of patients who participated in the experiment of the present invention. In Figure 4, numbers are expressed as mean ± standard deviation or n (%), where ApoE represents apolipoprotein E, cSS represents cortical superficial siderosis, and ICH represents intracerebral hemorrhage. Referring to Figure 4, 71 participants were recruited in the experiment of the present invention, of which 25 participants were Aβ negative (-) and the remaining 46 participants were Aβ positive (+).

Figure 5 is a graph showing the importance of variables in the GBM model and RF model in the experiment of the present invention. In Figure 5, HTN represents hypertension.

Referring to Figure 5, as important predictors for Aβ positivity in the present invention, the relative importance was calculated using the GBM and RF algorithms among 17 clinical and imaging variables, and the most important variables that were similar in the two algorithms were selected. The five important variables ranked and their relative importance in the GBM model are lobar CMB number (18.6), deep CMB number (8.8), lacunes number (5.7), age (4.6), and number (4.6). Meanwhile, the RF model has six important variables: lobar CMB number (60.4), deep CMB number (23.7), lacune number (23.3), age (15.4), absence of diabetes (8.4), and dentate nucleus CMB number (6.8). selected.

Figure 6 is a graph showing optimal threshold values detected at multiple change points in the GBM model and RF model in the experiment of the present invention. In Figure 6, ACC is accuracy, MCC is misclassification, Class ACC is class per accuracy, F1 is the harmonic mean of positive and negative prediction values with equal weights, and F0.5 gives more weight to PPV than NPV. Represents the average of semi-positive and negative predicted values.

In Figure 6, as the cutoff value of the predictor for Aβ positivity, the threshold in GBM was determined to be 0.7043 when each of the four metrics (F0.5, ACC, MCC, class ACC) is at its maximum, and in RF it is 3. The threshold was determined to be 0.6561 when each of the metrics (F1, accuracy, and misclassification) was at its maximum value.

In the experiment of the present invention, the cutoff values of important variables were determined using the threshold values obtained as above.

Figure 7 is a chart showing cut-off values in the GBM model and RF model in the experiment of the present invention, and Figure 8 shows the cut-off values of important variables in the GBM model and RF model in the experiment of the present invention. It's a graph.

Referring to Figures 7 and 8, the cutoff values of variables for predicting Aβ positivity are as follows. In GBM and RF, (1) lobar microbleeds (CMBs) greater than 14, (2) no deep CMBs, (3) number of lacunar cerebral infarcts less than or equal to 7, (4) age greater than 64, (5) This is a case where there is no microbleed (CMB) in the dentate nucleus.

Figure 9 is a chart summarizing the performance of the GBM model and the RF model in the experiment of the present invention. In Figure 9, the lower the value of the mean square error (MSE), root mean square error (RMSE), log loss, average error per class, and Gini impurity, the better the prediction ability, and the lower the SD, the higher the reliability.

As shown in Figure 9, the performance of GBM and RF-based prediction models was modeled in the experiment of the present invention, and as a result, both GBM and RF models showed good performance. MSE was 0.14 ± 0.02 in GBM and 0.18 ± 0.06 in RF. RMSE was 0.41 ± 0.08 in GBM and 0.37 ± 0.03 in RF. Logarithmic loss was 0.47 ± 0.07 for GBM and 0.53 ± 0.17 for RF. The mean per class error was 0.22 ± 0.06 for GBM and 0.25 ± 0.14 for RF. Gini impurity was 0.65 ± 0.09 in GBM and 0.60 ± 0.24 in RF. Area under curve (AUC) was 0.83 ± 0.04 in GBM and 0.80 ± 0.12 in RF. Precision-recall AUC was 0.86 ± 0.04 in GBM and 0.67 ± 0.18 in RF.

Although the present invention has been described above using several preferred examples, these examples are illustrative and not limiting. Those of ordinary skill in the technical field to which the present invention pertains will understand that various changes and modifications can be made without departing from the spirit of the present invention and the scope of rights set forth in the appended claims.

Claims

In the amyloid FET positive prediction method for patients suspected of amyloid angiopathy in the amyloid FET positive prediction device,

Receiving brain imaging data for the patient from the amyloid pet positive prediction device;

Performing machine learning on brain image data received from the amyloid pet positive prediction device; and

Predicting the amyloid PET positivity rate for the patient through the machine learning in the amyloid PET positivity prediction device.

Amyloid PET positive prediction method including.
In claim 1,

A method for predicting amyloid PET positivity, characterized in that the brain imaging data is brain MRI (magnetic resonance imaging) data.
In claim 1,

The machine learning is an amyloid PET positive prediction method, characterized in that the GBM (gradient boosting machine) algorithm.
In claim 1,

The machine learning is an amyloid PET positive prediction method, characterized in that the RF (random forest) algorithm.
In claim 3 or claim 4,

In order to predict the amyloid PET positivity rate, machine learning is performed using the number of lobar cerebral microbleeds, lacunes, patient's age, and microbleeds in the dentate nucleus as variables. Amyloid PET positive prediction method.
In the amyloid PET positive prediction device for patients suspected of amyloid angiopathy,

a receiving unit for receiving brain image data for the patient;

a prediction unit that performs machine learning on the brain image data received from the receiver and predicts an amyloid PET positivity rate for the patient through the machine learning; and

Display unit to display the amyloid pet positivity rate predicted in the prediction unit

Amyloid PET positive prediction device comprising a.
In claim 6,

An amyloid PET positive prediction device, characterized in that the brain imaging data is brain MRI (magnetic resonance imaging) data.
In claim 6,

An amyloid PET positive prediction device, characterized in that the machine learning is a GBM (gradient boosting machine) algorithm.
In claim 6,

An amyloid PET positive prediction device, characterized in that the machine learning is an RF (random forest) algorithm.
In claim 8 or claim 9,

In order to predict the amyloid PET positivity rate, the prediction unit performs machine learning using the number of lobar cerebral microbleeds, lacunes, patient's age, and microbleeds in the dentate nucleus as variables. Amyloid pet positive prediction device characterized in that.