Probability Density Function Tissue Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for combining imaging data from different modalities, such as MRI and MRS, face challenges in achieving accurate tissue type classification and spatial localization of brain tumors due to differences in spatial resolution, often requiring interpolation or resolution reduction, which compromises image quality and accuracy.
Innovation Solution
A computer-implemented method that uses probability density functions derived from reference imaging data to assign tissue types to voxels, allowing for effective combination of imaging data from different modalities without interpolation or statistical assumptions, enabling higher resolution tissue type classification and spatial localization.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If imaging data from different modalities (MRI and MRS) are combined using traditional methods, then tissue type classification can be performed, but spatial resolution is degraded due to interpolation or resolution reduction
Solution Approach 1:
The patent changes the fundamental parameter of how multi-modal imaging data is integrated by using a random forest classifier that processes features from both MRI and MRS modalities independently at their native resolutions, then combines classification results through probability integration rather than spatial interpolation. This allows maintaining the high spatial resolution of MRI while incorporating the tissue characterization capabilities of MRS.
Solution Approach 2:
The patent segments the classification process into distinct stages: (1) extracting features from MRI data, (2) extracting features from MRS data, (3) independently classifying each modality's data, and (4) integrating results through probability combination. This segmentation allows each modality to contribute at its optimal resolution without requiring spatial interpolation between them.
2Measurement precision
If MRS data is used for tissue classification, then metabolic information accuracy is improved, but spatial localization precision deteriorates due to low spatial resolution requiring heavy image interpolation
Solution Approach 1:
The patent adds a probabilistic dimension to the spatial data by converting MRS classification results into probability distributions that are integrated with MRI-based spatial localization. This allows the metabolic information from MRS to enhance tissue classification accuracy without requiring spatial interpolation, as the probabilistic framework naturally handles the different spatial resolutions of the two modalities.
Solution Approach 2:
The patent introduces a random forest classifier as an intermediary that processes features from both MRI and MRS modalities. This intermediary extracts relevant features from each modality independently, classifies them separately, and then integrates the results through probability combination, thereby preserving the high spatial resolution of MRI while incorporating the metabolic accuracy of MRS.
3Measurement precision
If biopsy is performed for gold standard classification, then diagnostic accuracy is improved, but patient risk and procedure complexity increase
Solution Approach 1:
The patent creates a non-invasive diagnostic copy of the biopsy process by using a random forest classifier trained on multi-modal imaging features (MRI and MRS) to simulate histopathological classification. This virtual biopsy achieves comparable diagnostic accuracy for tumor grading without the physical risks, morbidity, and limitations of actual biopsy procedures such as sampling error and invasive risks.
Data Source
Figure 1
Figure 2
Figure 3(a)~3(c)
AI summary
Methods and apparatus for obtaining probability density functions representing expected distributions of values of a parameter associated with an imaging modality are disclosed. The probability density functions are derived using data obtained from reference tissue volumes using the same imaging modality and at least one other type of imaging modality. The probability density functions are used to analyze data obtained from a volume of tissue of a patient in order to classify the tissue according to tissue type. Methods and apparatus are also disclosed in which deviations from a mean of an arctangent of ratios of first and second metabolite intensities in voxels are used to identify tissue types.