WO2010085856A1

WO2010085856A1 - Brain disease progression scoring method and apparatus

Info

Publication number: WO2010085856A1
Application number: PCT/AU2010/000095
Authority: WO
Inventors: Jurgen Fripp; Olivier Salvado; Pierrick Bourgeat
Original assignee: Commonwealth Scientific and Industrial Research Organization CSIRO
Current assignee: Commonwealth Scientific and Industrial Research Organization CSIRO
Priority date: 2009-02-02
Filing date: 2010-02-02
Publication date: 2010-08-05
Anticipated expiration: 2011-08-02

Abstract

A method of characterising the state of brain disease within a candidate subject, the method comprising the steps of: (a) acquiring brain data of the brain of a number of subjects, each of the subjects having a known brain state or condition relative to the brain disease; (b) utilising a dimensionality reduction statistical method on the brain data to form a series of component vectors, (c) collecting like brain data states to determine a healthy brain cluster position in the principal component space and a brain disease condition cluster position in the principal component space; (d) for the candidate subject, acquiring the physiological brain data for the candidate subject, and, mapping the candidate subject's brain data to the components vectors; (e) determining a relative position of the mapping of the candidate subject's brain data relative to the healthy cluster position and the brain condition cluster position; and utilising the relative position as a measure of characterisation of the state of brain disease within the candidate subject.

Description

Brain Disease Progression Scoring Method and Apparatus

Field of the Invention

[0001] The present invention relates to the field of Biomedical Image Processing, and, in particular, discloses a method and apparatus for scoring the progression of a brain disease such as Alzheimer's disease.

Background

[0002] Neurodegenerative diseases such as Alzheimer's disease cause untold damage to the individual and surrounding societal structures. A brain disease, such as Alzheimer's disease is characterised by the accumulation in the brain of Amyloid plaques, leading to neuronal death, and cognitive decline. The affects of Alzheimer's disease normally result in a slow progression over many years. Clinicians usually categorized a subject as Healthy control (HC), Mild cognitive impaired (MCI), and Alzheimer's disease (AD), based on cognitive tests. Patients clinically diagnosed with MCI have been found to convert to AD at a rate of up to 30% over three years. [0003] Several imaging modalities are known to show different aspects of the disease: For example:

[0004] 1) PET-PiB shows accumulation of plaques (new marker);

[0005] 2) MRI-TlW shows loss of tissue (atrophy);

[0006] 3) PET-FDG shows loss of neuronal activity; and more.

[0007] Classification of a patient as HC, MCI or AD can be difficult and prone to misclassification. Further, at some point in time some progression from HC to AD is seen in most individual's life.

[0008] It has been shown that, through the use of PET-PiB, plaques accumulate and spread in the brain with a characteristic pattern. It has also been shown that the brain's neurons die with a different specific pattern. Other modalities reveal other patterns of evolution of this disease. [0009] Currently, most clinical use of PET-PiB involves the simple calculation of signal average in some part of the brain (e.g. neocortex) after normalization with some other parts (e.g. cerebellum). This is usually converted into a binary diagnosis of PiB+ or PiB- by thresholding at some specific value (i.e. PiB+ if the average signal of the neocortex is greater than 1.5 times the average signal of the cerebellum). [0010] Currently, there is only limited sensitivity and change observed in patients when using a standard uptake value ratio (SUVR) to evaluate longitudinal progression. Direct analysis using existing methods such as SUVR or hippocampus volume does not provide a sensitive enough measure to allow longitudinal changes to be detected in short time periods and relate them to clinical decisions.

[0011] Most methods classify patients as HC or AD, with the MCI cases observed to be either HC or AD-like. These methods have a number of problems, including: Limited sensitivity of SUVR to longitudinal changes; Quantification of the distribution of each pattern requires complex statistical analysis; Most studies show differences in population, but few allow quantifying of an individual; Each individual needs to be evaluated in term of his progression towards AD, possibly taking into account all the information available (imaging modalities, but also demographics, genotyping, etc). Summary of the Invention

[0012] It is an object of the present invention to provide an improved brain disease progression scoring method and apparatus.

[0013] In accordance with a first aspect of the present invention, there is provided a method of characterizing the state of brain disease within a candidate subject, the method comprising the steps of: (a) acquiring brain data of the brain of a number of subjects, each of the subjects having a known brain state or condition relative to the brain disease; (b) utilizing a statistical method to map the brain data to a reduced dimensional space; (c) collecting like brain data states to determine a healthy brain cluster position in the reduced dimensional space and a brain disease condition cluster position in the reduced dimensional space; (d) for the candidate subject, acquiring the brain data for the candidate subject, and, mapping the candidate subject's brain data to the reduced dimensional space; and (e) determining a relative position of the mapping of the candidate subject's brain data relative to the healthy cluster position and the brain condition cluster position; and utilizing the relative position as a measure of characterization of the state of brain disease within the candidate subject. [0014] In accordance with a further aspect of the present invention, there is provided a method of characterising the state of brain disease within a candidate subject, the method comprising the steps of: (a) determining physiological brain data of the brain of a number of subjects, each of the subjects having a known brain state or condition relative to the brain disease; (b) utilising a statistical analysis (e.g. principal component analysis) process on the brain data to reduce the large number of pixels to form a series of component vectors, (c) collecting like brain data states to determine a healthy brain cluster position in the component space and a brain disease condition cluster position in the component space; (d) for the candidate subject, acquiring the brain data for the candidate subject, and, mapping the candidate subject's brain data to the components vectors; (e) determining a relative position of the mapping of the candidate subject's brain data relative to the healthy cluster position and the brain condition cluster position; and utilising the relative position as a measure of characterisation of the state of brain disease within the candidate subject. [0015] In accordance with a further aspect of the present invention, there is provided a system for characterizing the state of a brain disease within a candidate subject, the system including: a first acquisition unit for acquiring first brain data from a first group of subjects having a known brain state or condition relative to the brain disease; a data dimensionality reduction unit for statistically reducing the first brain data to a corresponding reduced dimensional space; a first processing unit for processing the reduced first brain data to determine a healthy brain cluster position in the reduced dimensional space and a brain disease condition cluster position in the reduced dimensional space; a further acquisition unit for acquiring corresponding brain data for said candidate subject; a further data dimensionality reduction unit for dimensionally reducing the candidate subject brain data to corresponding candidate reduced brain data; and a further processing unit for determining a relative position of the candidate reduced brain data to the healthy brain cluster position and the brain disease condition cluster position, and outputting the relative position information as a state of a brain disease within a candidate subject. [0016] Preferably, the physiological brain data can include at least one of a PET-PiB scan of the brain or a MRI scan of the brain. In some embodiments, the brain disease can be Alzheimer's disease. The dimensionality reduction method can be principal component analysis, the number of principal components utilised can be two. The brain data can comprise at least two separate scans of a subject's brain using different scanning modalities. In some embodiments, the relative position of the mapping of the candidate subject's brain data can be determined in a non-linear manner.

Brief Description of the Drawings [0017] Benefits and advantages of the present invention will become apparent to those skilled in the art to which this invention relates from the subsequent description of exemplary embodiments and the appended claims, taken in conjunction with the accompanying drawings, in which: Fig. 1 is a flow chart of the steps of patient processing in the preferred embodiment; Fig. 2 is a flow chart of the steps involved in background preparation of data; Fig. 3 is a flow chart of the steps involved in calculating of HC and AD clusters;

Fig. 4 is a flow chart of calculating a patient's cluster position; Fig. 5 illustrates example plots of data points on two principal components; and - A -

Fig. 6 illustrates the process of determination of clusters and an associated composite score.

Detailed Description

[0018] Preferred embodiments of the invention will now be described, by way of example only, with reference to the accompanying drawings.

[0019] The preferred embodiment involves a software system that allows the scoring of the degree of Alzheimer's disease in a patient from their medical images (PET and/or MR). This is achieved by characterizing the amyloid load and neuronal loss suffered by the patient, which is compared to statistical analysis performed on a large population and (if available) previous scans. [0020] Using a model that characterizes the pattern of evolution rather than a simple signal average, allows the better characterization of the likely progression of HC, MCI and AD subjects in a fully automatic way.

[0021] This preferred embodiment provides a method used to characterize and score the medical images, and its use allows evaluating the presence or risk of developing Alzheimer's disease (AD) or related conditions. The preferred embodiments have practical use in the early diagnosis of disease, in monitoring humans at risk of developing AD, and in enabling better treatment and management decisions to be made in clinically and sub-clinically affected humans. [0022] The preferred embodiment provides a continuous value to characterize patients' progression. Previous approaches have had limited sensitivity and specificity in diagnosing AD. The present approach has been found to improve this significantly by increasing the differentiation between AD and PiB+ NCs. It is also more sensitive to longitudinal changes. [0023] The preferred embodiment provides for an implementation in the form of computer software program that can be used by physicians and researchers to characterize and quantify Alzheimer's disease for a subject or a group of subjects. [0024] Fig. 1 illustrates the steps 10 of the preferred embodiment. For each subject, the user loads 11 the scans from PET-PiB, MRI-TlW or any other available scan and provide information about the subject (e.g. age, gender). The scans are then utilised to produce a corresponding 3-D volume of the brain containing a series of voxels. The software then computes a score between 0.0 and 1.0. A 0.0 corresponds to a healthy person and 1.0 to advanced Alzheimer's disease, taking into account the demographics of the subject such as age, gender, etc

[0025] The scoring can then be used either to help in diagnosing the subject, to assess the efficacy of a treatment (the score should go down if the treatment is effective), or to compute the average score of a group of individuals in order to study a new therapy or a specific characteristic of the group (e.g. genetic mutation).

[0026] As a preliminary step, the scans from a large number of predefined subjects are loaded into the system. The large population of subjects has been previously clinically diagnosed as HC, MCI and AD (typically the AIBL database is utilised).

[0027] For each of the predefined subjects, a number of steps are taken as indicated 20 in Fig. 2. Each subject's scans are initially acquired 21 and entered into a template containing the patients demographics. Next the images are corrected for image artefacts and degradations 24, 25. Finally, a principle component analysis (PCA) is performed 26 on the voxel intensity values to allow a statistical analysis to extract the pattern of evolution of the disease for each imaging modality. For example, using only the PiB, the PCA allows for the computation of an axis that shows the pattern of evolution in the whole population from HC to AD.

[0028] The axis of evolution can be determined via the steps illustrated in Fig. 3. All known subjects are initially mapped to the first few principal components 31. Next, those in a healthy group (HC), having no symptoms or image features (e.g. no plaque detected by PiB) are collated and a "center location" determined. A second group of patients having an advanced stage of AD and with the typical image features (e.g. pronounced pattern of plaque deposition detected by PiB) is determined 32. A center location is also determined for this group. The average center location of each group defines two prototypes that we use to normalize the axis between 0 (HC) and 1 (AD) 33.

[0029] For a new individual, the steps of Fig. 4 are carried out. Initially, the scans are acquired and processed 42 in the usual way by alignment to a template and image corrections etc. Next, a corresponding position along the axis between HC and AD is computed giving a score between 0 and 1. This "AD score" reflect the progression of the subject towards AD. It provides a quantitative assessment of a new subject at a single time point, and allows monitoring the disease progression on a given subject, or a population.

[0030] In alternative embodiments, composite scores can be computed using patterns from different sources. Either using two different scores (e.g. plaque deposition and tissue loss), or using a single axis in a multi-dimensional scoring system. Further, the axis can be linear or non linear (e.g. a smooth manifold).

[0031] The preferred embodiment allows for a more accurately characterize of features of PiB data and relates them to clinical outcomes using a simple relationship to a trained model. As will be evident, the preferred embodiment utilizes training to estimate weighted images to characterize

AD progression.

[0032] Fig. 5 shows an initial set of example results. Statistical analysis of a population of subjects using the two first components of principal component analysis (PCA) plotted on each axis. Each subject is projected in the space defined by the two axis and is associated with a specific Amyloid plaque pattern (some examples are shown). The plot point shape codes the clinical assessment. A group of healthy examples is seen on the right, whereas the group of AD clusters on the left.

[0033] As shown in Fig. 6, once the HC and AD group are identified, a new axis defining a score is computed between HC (0.0) and AD (1.0). In this example, two dimensions are used corresponding to the two first principal components of a statistical analysis of PET-PiB images.

Alternatively, only one dimension could be used. Other dimension measurements could also be used such as the volume of gray matter as detected by MRI

[0034] Although the present invention has been described with particular reference to certain preferred embodiments thereof, variations and modifications of the present invention can be effected within the spirit and scope of the following claims.

Claims

We Claim:

1. A method of characterizing the state of brain disease within a candidate subject, the method comprising the steps of:

(a) acquiring brain data of the brain of a number of subjects, each of the subjects having a known brain state or condition relative to the brain disease;

(b) utilizing a statistical method to map the brain data to a reduced dimensional space;

(c) collecting like brain data states to determine a healthy brain cluster position in the reduced dimensional space and a brain disease condition cluster position in the reduced dimensional space; (d) for the candidate subject, acquiring the brain data for the candidate subject, and, mapping the candidate subject's brain data to the reduced dimensional space; (e) determining a relative position of the mapping of the candidate subject's brain data relative to the healthy cluster position and the brain condition cluster position; and utlising the relative position as a measure of characterisation of the state of brain disease within the candidate subject.

2. A method as claimed in claim 1 wherein the statistical method is a principal component analysis reduction of the brain data.

3. A method as claimed in claim 1 wherein the brain data includes a PET-PiB scan of the brain.

4. A method as claimed in any previous claim wherein the brain data includes a MRI scan of the brain.

5. A method as claimed in any previous claim wherein the brain disease is Alzheimer's disease.

6. A method as claimed in any previous claim wherein said brain data comprises at least two separate scans of a subject's brain using different scanning modalities.

7. A method as claimed in any previous claim wherein the relative position of the mapping of the candidate subject's brain data is determined in a non-linear manner.

8. A method of characterising the state of brain disease within a candidate subject, the method comprising the steps of:

(a) determining physiological brain data of the brain of a number of subjects, each of the subjects having a known brain state or condition relative to the brain disease; (b) utilising a statistical analysis process on the brain data to reduce the brain data to form a series of component vectors in a component space;

(c) collecting like brain data states to determine a healthy brain cluster position in the component space and a brain disease condition cluster position in the component space;

(d) for the candidate subject, acquiring the brain data for the candidate subject, and, mapping the candidate subject's brain data to the corresponding candidate subject components vectors; and

(e) determining a relative position of the mapping of the candidate subject's brain data relative to the healthy cluster position and the brain condition cluster position; and utilising the relative position as a measure of characterisation of the state of brain disease within the candidate subject.

9. A system for characterizing the state of a brain disease within a candidate subject, the system including: a first acquisition unit for acquiring first brain data from a first group of subjects having a known brain state or condition relative to the brain disease; a data dimensionality reduction unit for statistically reducing the first brain data to a corresponding reduced dimensional space; a first processing unit for processing the reduced first brain data to determine a healthy brain cluster position in the reduced dimensional space and a brain disease condition cluster position in the reduced dimensional space; a further acquisition unit for acquiring corresponding brain data for said candidate subject; a further data dimensionality reduction unit for dimensionally reducing the candidate subject brain data to corresponding candidate reduced brain data; and a further processing unit for determining a relative position of the candidate reduced brain data to the healthy brain cluster position and the brain disease condition cluster position, and outputting the relative position information as a state of a brain disease within a candidate subject.

10. A system when implementing the method of any of claims 1 to 8.

11. A method of characterising the state of brain disease within a candidate subject substantially as herein described with reference to any one of the embodiments of the invention illustrated in the accompanying drawings and/or examples.

12. A system of characterising the state of brain disease within a candidate subject substantially as herein described with reference to any one of the embodiments of the invention illustrated in the accompanying drawings and/or examples.