Biomarker for early detection of alzheimer disease

Abstract

The present disclosure relates to a method for providing biomarker for early detection of Alzheimer's Disease (AD), and particularly to a method that is able to enhance the accuracy of predicting AD from Mild Cognitive Impairment (MCI) patients using the Hippocampus magnetic resonance imaging (MRI) scans and Mini-Mental State Examination (MMSE) data. The providing MRI images containing the anatomical structure of Hippocampus biomarker and MMSE data as a training data set; training a processor using the training data set, and the training comprising acts of receiving MRI images and MMSE data as a testing data set from a target; and classifying the test data by the trained processor to include aggregating predictions.

Claims

1. A method of providing biomarker for early detection of Alzheimer's Disease, comprising: providing magnetic resonance imaging (MRI) images containing the anatomical structure of Hippocampus and Mini-Mental State Examination (MMSE) data as a training data set; training a processor using the training data set, and the training comprising acts of proceeding an MRI image preprocessing to determine volume of each Hippocampus in MRI images, and segmenting each Hippocampus into sections; determining surface areas for each section of each Hippocampus in MRI images; determining an average of Ratio of Principal Curvature (RPC) for each section of each Hippocampus, wherein the average of RPC for each section of each Hippocampus is a ratio of SUM(maximum curvature) and SUM(minimum curvature) for each section of each Hippocampus; and selecting candidate parameters as inputs to iteratively train an iterative neural network in the processor, wherein the candidate parameters comprise the volume of Hippocampus, the surface area and the average of the RPC for each section of Hippocampus, and scores of MMSE data; receiving MRI images and MMSE data as a testing data set from a target; and classifying the testing data set by the trained processor to include aggregating predictions.

2. The method as claimed in claim 1, wherein the act of proceeding an MRI image preprocessing to determine volume of each Hippocampus in MRI images and segmenting each Hippocampus into sections, comprises an intensity normalization, linear stereotaxic registration, creating linear mask, linear classification and linear segmentations.

3. The method as claimed in claim 2, wherein the segmentations of the Hippocampus are alveus, parasubiculum, presubiculum, subiculum, CA1, CA2/3, CA4, GC-DG, HATA, fimbria, molecular layer, Hippocampus fissure and Hippocampus tail.

4. The method as claimed in claim 3, wherein the candidate parameter comprises values of Hippocampus volume, Subiculum surface area, CA1 surface area, CA3 surface area, an average of RPC for Subiculum, an average of RPC for CA1, and an average of RPC for CA3.

5. The method as claimed in claim 1, wherein the candidate parameter comprises scores of orientations, attention, recall and language in the MMSE data.

6. The method as claimed in claim 1, wherein the act of classifying the MRI images and MMSE data by the trained processor are proceed in different iterative neural networks.

7. The method as claimed in claim 1, wherein the acts of preprocessing and segmenting the MRI images are performed by a computer using FreeSurfer.

8. The method as claimed in claim 1, wherein the act of determining surface areas for each section of each Hippocampus in MRI images comprises acts of reconstructing the each identified Hippocampus with sections to build a 3 Dimensions (3D) Hippocampus model, smoothing the surface area of the 3D Hippocampus model; and calculating the surface areas of each section of Hippocampus.

9. The method as claimed in claim 8, wherein the 3D Hippocampus model is built by Marching cubes.

10. The method as claimed in claim 8, wherein the act of smoothing the surface area of the 3D Hippocampus model is achieved by Laplacian smoothing.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

(1) Embodiments of the invention are described in more details hereinafter with reference to the drawings, in which:

(2) FIG. 1 is a flowchart illustrating a method for predicting Alzheimer's Disease (AD) in accordance with an embodiment of the present invention;

(3) FIG. 2 is an exemplary diagram illustrating the observing factors in the AD progression;

(4) FIG. 3A is a flowchart illustrating the acts of S111 with corresponding MRI images for each of steps in accordance with an embodiment of the present invention;

(5) FIG. 3B is an exemplary diagram illustrating the sections of the Hippocampus;

(6) FIG. 4A is an exemplary diagram of a 3D Hippocampus model in accordance with an embodiment of the present invention;

(7) FIG. 4B is a flowchart illustrating the step of S112 in accordance with an embodiment of the present invention; and

(8) FIG. 5 is an exemplary diagram of an iterative neural network in accordance with an embodiment of the present invention.

DETAILED DESCRIPTION

(9) In the following description, For simplicity and illustrative purposes, the present disclosure is described by referring mainly to an example thereof. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. It will be readily apparent however, that the present disclosure may be practiced without limitation to these specific details. In other instances, some methods and structures have not been described in detail so as not to unnecessarily obscure the present disclosure. As used herein, the terms “a” and “an” are intended to denote at least one of particular elements, the term “includes/comprises” means includes but not limited to, the term “including/comprising” means including but not limited to, and the term “based on” means based at least in part on.

(10) With reference to FIGS. 1 and 2, FIG. 1 is a flowchart illustrating a method for predicting AD in accordance with an embodiment of the present invention; and FIG. 2 is an exemplary diagram illustrating the observing factors in the AD progression. As shown in FIG. 2, the Hippocampus volume is considered the most important and crucial factor to distinguish individuals with AD, especially from the ones who have MCI is crucial within the realm of early detection of AD.

(11) The objective of the present invention is to create a predictive computing model of AD by considering detailed structural and anatomic information contained within the MRI images as well as cognitive function assessed using the MMSE.

(12) Accordingly, as shown in FIG. 1, in an embodiment, A method for predicting Alzheimer's Disease comprising acts of S100 providing MRI images containing the anatomical structure of Hippocampus and MMSE data as a training data set, S110 training a processor using the training data set, S120 receiving MRI images and MMSE data as a testing data set from a target, and S130 classifying the test data by the trained processor to include aggregating predictions.

(13) The acts of S110 training of the processor using the training data set is comprising acts of S111 proceeding an MRI image preprocessing to determine volume of a Hippocampus of each MRI images, and segmenting the Hippocampus into sections, S112 determining surface areas for each section of each Hippocampus in MRI images, S113 determining a Ratio of RPC for each section of each Hippocampus, and S114 selecting candidate parameters as inputs to iteratively train an iterative neural network in the processor. the candidate parameters are selected from the volume of Hippocampus, the surface areas and PRC of sections of Hippocampus, and scores of the MMSE data.

(14) With further reference to FIGS. 3A and 3B, FIG. 3A is a flowchart illustrating the acts of S111 with corresponding MRI images for each of steps in accordance with an embodiment of the present invention; and FIG. 3B is an exemplary diagram illustrating the sections of the Hippocampus. As shown in FIG. 3A, in this embodiment, the MRI image preprocessing S111 comprising acts of S1110 an intensity normalization, S1111 linear stereotaxic registration, S1112 creating linear mask, S1113 linear classification and S1114 linear segmentations.

(15) In the present disclosure, the acts of S1110 to S1114 for preprocessing and segmenting the MRI images is proceed by a computer using FreeSurfer. FreeSurfer is a software for the analysis and visualization of structural and functional neuroimaging data from cross-sectional or longitudinal studies. It is developed by the Laboratory for Computational Neuroimaging at the Athinoula A. Martinos Center for Biomedical Imaging at Massachusetts General Hospital. For structural MRI image, FreeSurfer provides a cortical and subcortical full processing pipeline describing as following:

(16) 1. Intensity correction, noise filtering, artifact correction, skull stripping and gray-white matter segmentation; The surface-based stream

(17) 2. Reconstruction of cortical surface models (gray-white boundary surface);

(18) 3. Nonlinear registration of the cortical surface of an individual with a stereotaxic atlas (MNI305 atlas); The volume-based stream

(19) 4. Labeling of cortical regions and subcortical regions;

(20) 5. Statistical analysis of group morphometry differences; and

(21) 6. Subfields of Hippocampus segmentation.

(22) Therefore, the volume and the sections of Hippocampus is able to obtained by using the MRI image through FreeSurfer. As shown in FIG. 3B, the sections of Hippocampus at least comprise alveus, parasubiculum, presubiculum, subiculum, CA1, CA2/3, CA4, GC-DG, HATA, fimbria, molecular layer, Hippocampus fissure and Hippocampus tail.

(23) With reference to FIGS. 4A and 4B, FIG. 4A is an exemplary diagram of a 3 Dimensions (3D) Hippocampus model in accordance with an embodiment of the present invention; and FIG. 4B is a flowchart illustrating the step of S112 in accordance with an embodiment of the present invention. In act of S112, the Hippocampus is built in a 3D Hippocampus model with each identical section. In this embodiment, the 3D Hippocampus model is built by Marching cubes. The “marching cubes” described here is the technique directing to a patent, U.S. Pat. No. 4,710,876 “System And Method For The Display Of Surface Structures Contained Within The Interior Region Of A Solid Body” by Harvey E. Cline and William E. Lorensen, issued on Dec. 1, 1987.

(24) As shown in FIG. 4A, a block of volumetric data is displayed, it determines interfaces between adjacent data values indicating a change in the measured value, and then models the surfaces with triangular elements having a vector normal to the surface at each of the vertices of the triangle.

(25) Laplacian smoothing is then applied to the 3D Hippocampus model which is configured for improving the quality of the triangulation while remaining faithful to the original surface geometry. The Hippocampus surface and each surface areas of the sections are obtained after the surface smoothing. Therefore, as shown in FIG. 4B, the act of 112 further comprises acts of S120 reconstructing the each identified Hippocampus with sections to build a 3D Hippocampus model, S1121 smoothing the surface area of the 3D Hippocampus model; and S122 calculating the surface areas of each sections of Hippocampus.

(26) In act of S113, a curvature analysis to the 3D Hippocampus model is proceeded to determine an average of a Ratio of Principle Curvature (RPC) for each section of each hippocampus. During the calculation of curvature, maximum curvatures and minimum curvatures of each sections are calculated, wherein the average of RPC is
RPC.sub.avg=SUM(maximum curvature)/SUM(minimum curvature)

(27) With reference to FIG. 5, FIG. 5 is an exemplary diagram of an iterative neural network in accordance with an embodiment of the present invention. As describe in above, the volume, average of curvature for each section, and surface of the Hippocampus are obtained. These values are considered to be candidate parameters as inputs for the iterative neural network. As shown in FIG. 5, there're multiple input nodes, at least one output node and multiple layers of neuron nodes between the input and the output.

(28) According to inventor's experiments, in an prefer embodiment, the values of Hippocampus volume, Subiculum surface area, CA1 surface area, CA3 surface area, and the average RPCs of Subiculum, CA1 and CA3 are selected to be the candidate parameters. In contrast, unlikely conventional studies only relied on the volume of the Hippocampus, the present invention uses curvatures to quantify the Hippocampus as features.

(29) Besides neurological examinations, in order to enhance the accuracy of prediction, especially for predicting AD from MCI patients, the present invention further uses physical/cognitive function assessed using MMSE data to train the iterative neural network. the parameters both in MRI images and MMSE data can be used in a single machine learning model, or separately into two individual models. As a person skilled in the art will realize that for separating models, the results can be combined later using the majority voting.

(30) The MMSE is a brief cognitive assessment tool commonly used to screen for dementia. The MMSE is composed of 11 major items forming a 30-point questionnaire with five different domains of cognition analyses. The five domains are (1) Orientation, contributing a maximum of 10 points, (2) Memory/Recall, contributing a maximum of 6 points, (3) Attention and calculation, as a measure of working memory, contributing a maximum of 5 points, (4) Language, contributing a maximum of 8 points, and (5) Design copying, contributing a maximum of 1 point. The 11 items are temporal orientation (5 points), spatial orientation (5 points), immediate memory (3 points), attention/concentration (5 points), delayed recall (3 points), naming (2 points), verbal repetition (1 points), verbal comprehension (3 points), writing (1 points), reading a sentence (1 points), and constructional praxis (1 points). In general, the MMSE was administered and scored by a medical doctor certified in internal-medicine with extensive dementia experience.

(31) Accordingly, the scores of MMSE data are used as candidate parameters for additional inputs. Based on inventor's experiment, orientation, attention, recall and language are the most effective features for accuracy.

(32) Below three tables shows the accuracy results of predicting AD with different inputs in iterative neural networks. In this embodiment, a basic Multilayer Perceptron (MLP) architecture is used with numerous of hidden layers. Table I uses only neuropsychological data (i.e., the MMSE score) as parameters. The accuracy is between 67.78% to 72.22%. Table II uses only neuroimaging features (i.e., the volume, surface area and RPC of the Hippocampus) as parameters. The accuracy is between 67.16% to 72.65%. Table III is combination of the neuropsychological data and neuroimaging feature data. The accuracy is between 75.12% to 75.86%.

(33) TABLE-US-00001 TABLE I Input feature Basic MLP architecture Neuro- Number Neuropsychological data Selection psychological of 2 hidden layers with 4 neurons method test Features Accuracy Sensitivity Specificity Univariate MMSE 4 72.22% 70.37% 74.07% selection Feature MMSE 2 67.78% 57.04% 78.52% importance

(34) TABLE-US-00002 TABLE II Input feature Basic MLP architecture Number Neuroimaging features Selection of 2 hidden layers with 12 neurons method Criteria features Accuracy Sensitivity Specificity Univariate p-value < 0.01 11 67.16% 70.00% 64.32% selection Univariate p-value < 0.05 19 71.17% 69.38% 72.96% selection Univariate p-value < 0.1 25 69.26% 70.99% 67.53% selection Feature Random forest 10 67.59% 68.02% 67.16% importance Top 10 Feature Random forest 15 71.91% 69.75% 74.07% importance Top 15 Feature Random forest 20 72.65% 71.48% 73.83% importance Top 20 Feature Random forest 25 71.54% 69.01% 74.07% importance Top 25

(35) TABLE-US-00003 TABLE III Input feature Basic MLP architecture Neuroimaging + Neuro- neuropsychological psychological data 2 hidden layers test (.sub.Univariate Number with 15 neurons Criteria .sub.selection) of feature Accuracy Sensitivity Specificity Univariate selection method p-value < 0.05 MMSE 23 75.86% 74.07% 77.65% p-value < 0.1 MMSE 29 73.95% 72.10% 75.80% Feature importance method (Random forest classifier) Random MMSE 19 75.12% 76.05% 74.20% forest Top 15 Random MMSE 24 75.56% 74.94% 76.17% forest Top 20

(36) Accordingly, the present invention combines the structural information derived from functional information and neuroimaging data that is derived from quantifying the anatomical structure of a Hippocampus which achieves better accuracy of predicting AD.

(37) The foregoing description of the present invention has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations will be apparent to the practitioner skilled in the art.

(38) The embodiments were chosen and described in order to best explain the principles of the invention and its practical application, thereby enabling others skilled in the art to understand the invention for various embodiments and with various modifications that are suited to the particular use contemplated.