JP7313165B2 - Alzheimer's Disease Survival Analyzer and Alzheimer's Disease Survival Analysis Program - Google Patents

Description

The present invention relates to an Alzheimer's disease survival analysis device and an Alzheimer's disease survival analysis program for estimating the probability of whether or not a subject will transition to Alzheimer's disease in several years.

In recent years, in addition to diagnosing whether or not a person has a disease, there is a demand for techniques for analyzing the possibility of having a disease. For example, it is known that Alzheimer's disease (AD), which progresses slowly and gradually affects daily life, can be greatly delayed if detected early and appropriate treatment is started. Therefore, if it is possible to know the probability of transition to Alzheimer's disease in a few years, it will be possible to take appropriate measures in advance to delay the progression, which is very useful.

Until now, it has been difficult to predict the transition to Alzheimer's disease at the individual level. For example, the elderly with Mild Cognitive Impairment (MCI), which is a pre-stage of Alzheimer's disease, is said to progress to Alzheimer's disease at an annual rate of about 15%, and about half in five years. When diagnosed with mild cognitive impairment, such predictions are possible as a group, but there are various individual types, such as a type that immediately transitions to AD even with mild cognitive impairment and a type that remains mild cognitive impairment. Therefore, it is necessary to estimate when and with what probability people will develop Alzheimer's disease on an individual level rather than on a population level.

For example, Patent Literature 1 discloses a method of acquiring signs of dementia including Alzheimer's disease based on diffusion MRI images.

Japanese Patent Publication No. 2018-511457

In Patent Document 1, pattern classification using diffusion MRI images can discover signs of the presence of cognitive impairment at a certain point in time, but it cannot be used to predict the progression to dementia several years later.

The present invention has been made in view of the above problems, and aims to provide an Alzheimer's disease survival analysis device and an Alzheimer's disease survival analysis program for estimating the probability of whether or not a subject will transition to Alzheimer's disease in several years.

An Alzheimer's disease survival analysis apparatus according to the present invention is an Alzheimer's disease survival analysis apparatus for performing survival analysis processing for predicting a transition probability of a subject to Alzheimer's disease, comprising: a gray matter volume map data acquisition unit for acquiring gray matter volume map data in the subject's brain based on a gray matter volume map data acquisition means; and a transition probability output unit that inputs at least each of the gray matter density data of the plurality of regions of the subject and outputs the transition probability of the subject to Alzheimer's disease based on a trained model that has been trained in advance regarding outputting the transition probability of the subject to Alzheimer's disease using at least the gray matter density data of the plurality of regions as an input parameter.

Further, in the Alzheimer's disease survival analysis apparatus according to the present invention, the transition probability output unit obtains the transition probability to Alzheimer's disease of the subject based on the learned model as a cumulative Weibull distribution and displays it in a graph.

Further, in the Alzheimer's disease survival analysis apparatus according to the present invention, the gray matter volume map data acquisition means is MRI (magnetic resonance imaging), and the gray matter volume map data is acquired by obtaining a T1-weighted image using MRI.

Furthermore, in the Alzheimer's disease survival analysis apparatus according to the present invention, the trained model is obtained as a result of learning a neural network so as to output a transition probability to Alzheimer's disease.

An Alzheimer's disease survival analysis program according to the present invention is an Alzheimer's disease survival analysis program for causing a computer to implement a survival analysis process for predicting the transition probability of a subject to Alzheimer's disease, wherein the computer is provided with a gray matter volume map data acquisition function for acquiring gray matter volume map data in the brain of the subject based on gray matter volume map data acquisition means, and a gray matter volume in each region when the brain structure of the subject is divided into a plurality of predetermined regions based on the gray matter volume map data. and a transition probability output function for outputting the transition probability to Alzheimer's disease of the subject using at least the gray matter volume data for each area of the plurality of areas as an input parameter.It is characterized by

According to the present invention, it is possible to estimate the long-term transition probability rather than the risk at a particular point in time. That is, the advantages of the Alzheimer's disease survival analysis apparatus according to the present invention are: (1) It is possible to estimate the absolute value of the transition probability for each individual rather than the relative risk of the individual, (2) The accuracy does not decrease even in long-term estimation, and (3) It is optimized for gray matter volume map data.

1 is a block diagram showing the configuration of an Alzheimer's disease survival analyzer according to the present invention; FIG. 1 is a block diagram showing the configuration of a system including an Alzheimer's disease survival analyzer according to the present invention; FIG. 1 is a conceptual diagram showing the concept of dividing a brain structure into multiple regions; FIG. FIG. 4 is an explanatory diagram showing the concept of a trained model for outputting a transition probability to Alzheimer's disease; FIG. 4 is a flow chart diagram showing the flow of learning processing of a trained model used in the Alzheimer's disease survival analysis apparatus according to the present invention. FIG. 4 is a flow chart diagram showing the flow of survival analysis processing in the Alzheimer's disease survival analyzer according to the present invention. FIG. 10 is an explanatory diagram showing a display example when displaying and outputting a transition probability to Alzheimer's disease as a graph;

An example of the Alzheimer's disease survival analyzer according to the first embodiment will be described below with reference to the drawings. FIG. 1 is a block diagram showing the configuration of an Alzheimer's disease survival analyzer 10 according to the present invention. As shown in FIG. 1, an Alzheimer's disease survival analyzer 10 according to the present invention includes at least a gray matter volume map data acquisition unit 11, a regional gray matter volume data generation unit 12, a transition probability output unit 13, and a storage unit 14.

The Alzheimer's disease survival analysis apparatus 10 may be realized in one terminal device and used offline, or the functions of the Alzheimer's disease survival analysis apparatus 10 may be aggregated in a server device. FIG. 2 is a block diagram showing the configuration of a system including an Alzheimer's disease survival analyzer according to the present invention. In FIG. 2, 20 is a server device, and functions of the Alzheimer's disease survival analysis device 10 are aggregated in this server device 20 . The terminal devices 301 to 30n (n is an arbitrary integer) used by the user are each connected to the server device 20 via the communication network 40 such as the Internet, and the system may use the functions of the Alzheimer's disease survival analysis device 10. The server device 20 is managed by a system administrator and has various functions for providing information on various processes to the plurality of terminal devices 301 to 30n. In this example, the server device 20 is configured by an information processing device such as a WWW server, and includes a storage medium for storing various information. Note that the server device 20 has a general configuration for performing various processes as a computer, such as a control unit and a communication unit, but a description thereof will be omitted here. The system configuration is not limited to the example of FIG. 2, and may be configured such that one terminal device that functions as the Alzheimer's disease survival analysis device 10 is used by a plurality of users, or that a plurality of server devices are provided.

Each of the plurality of terminal devices 301 to 30n is connected to the communication network 40 and has hardware and software for executing various processes by communicating with the server device 20. FIG. Each of the plurality of terminal devices 301 to 30n may be configured to directly communicate with each other without going through the server device 20. For example, the terminal device 301 functioning as the Alzheimer's disease survival analysis device 10 may be accessed and used by direct communication from the other terminal devices 302 to 30n.

The gray matter volume map data acquisition unit 11 has a function of acquiring gray matter volume map data in the brain of the subject based on the gray matter volume map data acquisition means. Here, the gray matter volume map data refers to a three-dimensional map representing gray matter locations in the subject's brain. Further, the gray matter volume map data acquisition means refers to any means capable of acquiring gray matter volume map data. That is, any means may be used as long as it is possible to distinguish and recognize gray matter locations from other brain tissue locations containing white matter by performing brain tissue separation, identify gray matter locations in the entire brain, and obtain gray matter volume map data. Examples of means for obtaining gray matter volume map data include obtaining T1-weighted images using MRI (magnetic resonance imaging) and obtaining gray matter volume map data by specifying gray matter locations in brain tissue.

The region-specific gray matter volume data generation unit 12 has a function of generating region-specific gray matter volume data indicating the gray matter volume in each region when the subject's brain structure is divided into a plurality of predetermined regions based on the gray matter volume map data. Here, the predetermined plurality of regions refers to a plurality of regions obtained by dividing the brain structure into a plurality of regions according to a predetermined division rule. FIG. 3 is a conceptual diagram showing the concept of dividing the brain structure into multiple regions. As shown in FIG. 3, the brain is divided into a plurality of regions, and regional gray matter volume data is generated for each divided region. Although FIG. 3 shows divided regions that are visible when the side of the brain is displayed, since the three-dimensional structure is divided into a plurality of regions, regions that cannot be seen are also formed. The method of division can be set in various ways, but as standard methods for dividing the brain structure, examples of division into 116 regions by a method called Automated Anatomical Labeling (AAL) and division into 246 regions by a method called Brain Connectome Atlas (BA) are conceivable, but are not limited to these. To generate gray matter density data for each region, in each region, the gray matter volume within the region is obtained from the gray matter volume map data to generate region-specific gray matter volume data. That is, when the BA method is used to divide the area into 246 areas, 246 areas of gray matter volume data are generated for one subject. Each area-specific gray matter volume data is stored in a state of being associated with the corresponding relationship with the divided area.

The transition probability output unit 13 has a function of inputting at least the gray matter volume data of each region of the subject and outputting the transition probability of the subject to Alzheimer's disease, based on a pre-learned model for outputting the transition probability of the subject to Alzheimer's disease using at least the gray matter volume data of each region of the plurality of regions as an input parameter. FIG. 4 is an explanatory diagram showing the concept of a trained model for outputting a transition probability to Alzheimer's disease. As shown in FIG. 4, the trained model is configured by, for example, a neural network, and includes gray matter volume data for each region of multiple regions at least as an input parameter for the input layer. The input parameters for the input layer of the neural network must include at least regional gray matter volume data, but any parameter (+α parameter in FIG. 4) other than regional gray matter volume data may be adopted. For example, the age of the subject, the result of MMSE, which is a cognitive function score, the value of biomarkers such as amyloid β and tau protein in the cerebrospinal fluid and the amount of abnormal protein accumulation in the brain, various examinations such as blood test results of the subject's health checkup, including the subject's medical history regarding other diseases, etc. Various information can be adopted as input parameters. A learned model is obtained by learning a neural network so as to output the transition probability of the subject to Alzheimer's disease by learning processing described later. Any method may be used as long as it can output the transition probability to Alzheimer's disease. For example, a method of directly estimating the transition probability at each point in time after a specified number of years may be used. Alternatively, a neural network may be configured to estimate and output transition probabilities for each month and year since the subject's gray matter volume map data was acquired, such as AD transition probability after one year, AD transition probability after two years, ..., AD transition probability after five years, ..., AD transition probability after ten years, and so on. However, it is more preferable to add a restriction that the probability of transition to Alzheimer's disease follows the cumulative Weibull distribution, and to estimate the parameters of the Weibull distribution using a neural network. In the case of a method that directly estimates the transition probability at each time point, there is a possibility that the accuracy of the data including censored data will drop in the second half of the follow-up period.

When a neural network outputs transition probabilities as a Weibull distribution, for example, a configuration in which the parameters m and s in the function of the Weibull distribution {1-exp(-t ^m /s)} are output to the neural network is conceivable. In addition, the configuration of the input layer and the hidden layer of the neural network can be set in various ways. For example, when the brain structure is divided into 246 regions, the number of nodes in the input layer is 246 + α, the number of hidden layers is 3, and the number of nodes is 128, 64, and 32, respectively, and the output layer is configured to output two parameters m and s in the function of the Weibull distribution.

Further, the transition probability output unit 13 may have a function of obtaining the transition probability of the subject to Alzheimer's disease as a cumulative Weibull distribution based on the learned model and displaying it in a graph. That is, it is possible to output only the transition probability at each time point in response to an inquiry about the transition probability at each time point, such as "What is the transition probability after 5 years?", but by obtaining the transition probability of the subject as a cumulative Weibull distribution and displaying it in a graph, there is an effect that it is possible to visually recognize how much the probability of transition to Alzheimer's disease increases as the years pass.

The storage unit 14 has a function of storing various types of information such as information used in each unit of the Alzheimer's disease survival analysis apparatus 10 and information obtained as a result of processing by each unit. Also, the learned model used by the transition probability output unit 13 may be stored in the storage unit 14 .

FIG. 5 is a flow chart showing the flow of learning processing of a trained model used in the Alzheimer's disease survival analyzer according to the present invention. As shown in FIG. 5, the learning process of the neural network is started by acquiring sample data, which is a set of gray matter volume map data and data on the subsequent transition to Alzheimer's disease, in the Alzheimer's disease survival analyzer 10 (step S101). That is, sample data is obtained as a set of gray matter volume map data of a certain individual and data of years and months when the individual transitioned to Alzheimer's disease from the date when the gray matter volume map data was measured (or data indicating no transition). Next, the Alzheimer's disease survival analysis apparatus 10 generates region-specific gray matter volume data for each region when the brain structure is divided into a plurality of predetermined regions from the gray matter volume map data among the acquired sample data (step S102). In this step, regional gray matter volume data are generated for the number of regions. Then, the region-specific gray matter volume data is input to the learning target neural network, and the transition probability to Alzheimer's disease is output (step S103). Finally, the loss is calculated based on the output transition probability and the data of the year and month when the target individual transitioned to Alzheimer's disease, and the weights, biases, etc. of the neural network are updated so as to reduce the loss (step S104), and the process ends. Any method can be used to set the loss function, but for example, a method of preparing correct data with 0 before the individual transitions to Alzheimer's disease and 1 after the transition from sample data information and calculating the loss by the least squares method based on the difference between the correct data and the output transition probability can be considered. Although the flowchart shown in FIG. 5 only describes the learning process based on one set of sample data, in reality, a highly accurate trained model can be obtained by repeatedly performing the learning process based on multiple sets of sample data. Note that instead of executing the learning process in the Alzheimer's disease survival analysis apparatus 10, a trained model obtained by executing the learning process in another computer may be stored in the Alzheimer's disease survival analysis apparatus 10.

Next, the flow of survival analysis processing in the Alzheimer's disease survival analyzer will be described. FIG. 6 is a flow chart showing the flow of survival analysis processing in the Alzheimer's disease survival analyzer according to the present invention. In FIG. 6, survival analysis processing in the Alzheimer's disease survival analysis apparatus 10 is started by obtaining gray matter volume map data of the subject in the Alzheimer's disease survival analysis apparatus 10 (step S201). Next, the Alzheimer's disease survival analysis apparatus 10 generates regional gray matter volume data for each region when the brain structure is divided into a plurality of predetermined regions from the obtained gray matter volume map data of the subject (step S202). Then, the Alzheimer's disease survival analysis apparatus 10 inputs the region-specific gray matter volume data generated for each of the plurality of regions to the learned model, outputs the transition probability to Alzheimer's disease (step S203), and ends the survival analysis process.

FIG. 7 is an explanatory diagram showing a display example when the probability of transition to Alzheimer's disease is graphically displayed and output. As shown in FIG. 7, by obtaining the transition probability of the subject as a cumulative Weibull distribution and displaying it graphically, it is possible to visually recognize how much the probability of transition to Alzheimer's disease increases with the passage of years. In addition to the graphical display of the transition probability of the subject, for example, a typical example of a Weibull distribution of a person who did not transition to Alzheimer's disease within 5 years and a typical Weibull distribution of a person who transitioned to Alzheimer's disease within 5 years may be displayed together. By displaying these typical examples together, there is an effect that the subject's risk of transition to Alzheimer's disease becomes easy to visually understand.

As described above, according to the Alzheimer's disease survival analysis apparatus according to the present invention, the gray matter volume map data in the brain of the subject is obtained based on the gray matter volume map data obtaining means, the gray matter volume data for each region is generated based on the gray matter volume map data and indicates the gray matter volume in each region when the brain structure of the subject is divided into a plurality of predetermined regions, and the gray matter volume data for each region of the plurality of regions is used as at least an input parameter to output the transition probability to Alzheimer's disease in advance. Based on the trained model that has been learned, at least the gray matter volume data for each region of multiple regions of the subject is input, and the transition probability of the subject to Alzheimer's disease is output. Therefore, it is possible to estimate the long-term transition probability instead of the risk at a specific point in time. That is, the advantages of the Alzheimer's disease survival analysis apparatus according to the present invention are: (1) It is possible to estimate the absolute value of the transition probability for each individual rather than the relative risk of the individual, (2) The accuracy does not decrease even in long-term estimation, and (3) It is optimized for gray matter volume map data.

In the first embodiment, the Alzheimer's disease survival analysis apparatus 10 alone has been described as performing various types of processing, but the present invention is not limited to this. For example, the learned model may be stored in a server device 20 connectable from the Alzheimer's disease survival analysis apparatus 10 via a communication network, and the server device 20 may execute the transition probability output process. That is, the subject's gray matter volume map data acquired by the Alzheimer's disease survival analysis apparatus 10 or the region-specific gray matter volume data generated by the Alzheimer's disease survival analysis device 10 is transmitted to the server device 20 via a communication network, the server device 20 is provided with the same function as the transition probability output unit 13 (and may also be provided with the function of the region-specific gray matter volume data generation unit 12), the server device 20 executes the transition probability output process, and the transition output by the server device 20. Even if the Alzheimer's disease survival analysis apparatus 10 receives the probabilities via a communication network, it is possible to realize the Alzheimer's disease survival analysis apparatus 10 having the same effects as those of the above embodiments.

Needless to say, in the first embodiment described above, the Alzheimer's disease survival analysis device 10, the server device 20, and the plurality of terminal devices 301 to 30n execute the various processes described above according to various control programs stored in their own storage devices.

10 Alzheimer's Disease Survival Analyzer 11 Gray Matter Volume Map Data Acquisition Unit 12 Region-Specific Gray Matter Volume Data Generation Unit 13 Transition Probability Output Unit 14 Storage Unit 20 Server Device 301 to 30n Terminal Device 40 Communication Network

Claims (3)

An Alzheimer's disease survival analysis device for performing survival analysis processing for predicting a transition probability of a subject to Alzheimer's disease,
a gray matter volume map data acquisition unit that acquires gray matter volume map data in the subject's brain based on the gray matter volume map data acquisition means;
a gray matter volume data generation unit for generating region-specific gray matter volume data indicating the gray matter volume in each region when the subject's brain structure is divided into a plurality of predetermined regions based on the gray matter volume map data;
a transition probability output unit that inputs at least the gray matter volume data for each of the plurality of regions of the subject and outputs the transition probability of the subject to Alzheimer's disease, based on a pre-learned model for outputting the transition probability to Alzheimer's disease of the subject using at least the gray matter volume data for each of the plurality of regions as an input parameter;
The Alzheimer's disease survival analyzer, wherein the transition probability output unit is configured to obtain the transition probability of the subject to Alzheimer's disease as a cumulative Weibull distribution based on a trained model obtained as a result of performing learning of a neural network so as to output the transition probability to Alzheimer's disease, and to display the result in a graph. The Alzheimer's disease survival analysis apparatus according to claim 1 , wherein the gray matter volume map data acquisition means is MRI (magnetic resonance imaging), and gray matter volume map data is acquired by obtaining a T1-weighted image using MRI. An Alzheimer's disease survival analysis program for causing a computer to realize a survival analysis process for predicting a transition probability to Alzheimer's disease of a subject,
to said computer;
a gray matter volume map data acquisition function for acquiring gray matter volume map data in the subject's brain based on the gray matter volume map data acquisition means;
a gray matter volume data generating function for generating region-specific gray matter volume data indicating the gray matter volume in each region when the subject's brain structure is divided into a plurality of predetermined regions based on the gray matter volume map data;
Alzheimer's disease survival analysis program for realizing an Alzheimer's disease survival analysis program for realizing a transition probability output function of inputting at least the gray matter volume data of each region of the subject and obtaining the transition probability of the subject to Alzheimer's disease as a cumulative Weibull distribution and graphically displaying the transition probability of the subject to Alzheimer's disease by using at least the gray matter volume data of each region of the plurality of regions as an input parameter and outputting the transition probability to Alzheimer's disease of the subject, based on a trained model trained in advance by a neural network.

Images