FDG-PET Dementia Subtype Classification Using Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for diagnosing Alzheimer's disease (AD) and mixed dementia, such as visual interpretation of FDG-PET images, are subjective, time-consuming, and prone to inaccuracies due to the coexistence and heterogeneity of brain pathologies, making it difficult to distinguish between typical AD and mixed pathology.
Innovation Solution
A deep learning model, specifically a convolutional neural network (CNN) trained using brain imaging data, classifies patients between subtypes of dementia by analyzing metabolic patterns in brain regions, utilizing transfer learning and focusing on temporo-parietal hypo-metabolism and other regions, to provide accurate diagnoses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If visual interpretation of FDG-PET images is used to diagnose dementia subtypes, then diagnostic capability is provided, but the process is time-consuming and requires intensive training of expert staff
Solution Approach 1:
The patent applies automated image analysis that copies and processes FDG-PET images through computer algorithms instead of requiring manual visual interpretation by experts. The system automatically extracts metabolic patterns and classifies dementia subtypes, eliminating the time-consuming manual review process while maintaining diagnostic reliability.
Solution Approach 2:
The patent replaces the mechanical process of human visual interpretation with an automated computational system. The mechanical action of experts manually analyzing images is substituted by digital image processing algorithms that automatically detect hypo-metabolic regions and classify pathology types, significantly reducing time requirements.
2Reliability
If visual interpretation of FDG-PET images is used to diagnose dementia subtypes, then diagnostic capability is provided, but it is subjective and dependent on expertise
Solution Approach 1:
The system creates an objective digital copy of the diagnostic process by using automated algorithms that consistently apply the same analysis criteria to all images. This eliminates the subjectivity inherent in human interpretation, as the computational model provides reproducible, consistent classifications based on quantifiable metabolic patterns rather than expert opinion.
Solution Approach 2:
The patent transforms subjective visual assessment into objective quantitative parameters by measuring metabolic activity levels, hypo-metabolic region sizes, and spatial distributions. These numerical parameters provide precise, reproducible data that eliminates subjectivity and enables consistent differentiation between dementia subtypes based on measurable biological characteristics.
3Measurement precision
If semi-quantitative image analysis is used to detect hypo-metabolism, then sensitivity and specificity are improved, but results can be inaccurate when assessing small or adjacent regions
Solution Approach 1:
The patent transitions from two-dimensional semi-quantitative scoring to three-dimensional volumetric analysis by processing entire FDG-PET image volumes. The system analyzes metabolic patterns across multiple spatial dimensions and depth levels, enabling accurate assessment of small hypo-metabolic regions and their spatial relationships without the limitations of surface-level scoring methods.
Solution Approach 2:
The system segments the brain into multiple anatomical regions and further divides them into sub-regions to precisely locate and characterize hypo-metabolic areas. By segmenting the imaging data into manageable spatial units, the patent can accurately assess small regions and adjacent areas that would be difficult to evaluate using conventional semi-quantitative methods, improving both sensitivity and accuracy.
Data Source
AI summary
Diagnosis of dementia A method for diagnosing dementia subtypes is described. The method comprises obtaining brain imaging data relating to one or more patients, analysing the data using a deep learning model and classifying the one or more patients between a plurality of classes comprising a first class of patients having a first subtype of dementia and a second class of patients having a second subtype of dementia, using the deep learning model. The deep learning model has been trained using brain imaging data from patients, the brain imaging data comprising a first set of images showing evidence of temporo-parietal hypo-metabolism, and a second set of images showing evidence of hypo-metabolism in regions of the brain other than the temporal and preictal regions instead or in addition to the temporal and parietal regions, wherein the first set of images is labeled as associated with the first sub-type of dementia and the second set of images is labeled as associated with the second subtype of dementia. Related systems and methods are also described.