Refine
Document Type
- Article (2)
Language
- English (2)
Has Fulltext
- yes (2)
Is part of the Bibliography
- no (2)
Keywords
Institute
- Medizin (2)
The identification of pathological atrophy in MRI scans requires specialized training, which is scarce outside dedicated centers. We sought to investigate the clinical usefulness of computer-generated representations of local grey matter (GM) loss or increased volume of cerebral fluids (CSF) as normalized deviations (z-scores) from healthy aging to either aid human visual readings or directly detect pathological atrophy.
Two experienced neuroradiologists rated atrophy in 30 patients with Alzheimer's disease (AD), 30 patients with frontotemporal dementia (FTD), 30 with dementia due to Lewy-body disease (LBD) and 30 healthy controls (HC) on a three-point scale in 10 anatomical regions as reference gold standard. Seven raters, varying in their experience with MRI diagnostics rated all cases on the same scale once with and once without computer-generated volume deviation maps that were overlaid on anatomical slices. In addition, we investigated the predictive value of the computer generated deviation maps on their own for the detection of atrophy as identified by the gold standard raters.
Inter and intra-rater agreements of the two gold standard raters were substantial (Cohen's kappa κ > 0.62). The intra-rater agreement of the other raters ranged from fair (κ = 0.37) to substantial (κ = 0.72) and improved on average by 0.13 (0.57 < κ < 0.87) when volume deviation maps were displayed. The seven other raters showed good agreement with the gold standard in regions including the hippocampus but agreement was substantially lower in e.g. the parietal cortex and did not improve with the display of atrophy scores. Rating speed increased over the course of the study and irrespective of the presentation of voxel-wise deviations.
Automatically detected large deviations of local volume were consistently associated with gold standard atrophy reading as shown by an area under the receiver operator characteristic of up to 0.95 for the hippocampus region. When applying these test characteristics to prevalences typically found in a memory clinic, we observed a positive or negative predictive value close to or above 0.9 in the hippocampus for almost all of the expected cases. The volume deviation maps derived from CSF volume increase were generally better in detecting atrophy.
Our study demonstrates an agreement of visual ratings among non-experts not further increased by displaying, region-specific deviations of volume. The high predictive value of computer generated local deviations independent from human interaction and the consistent advantages of CSF-over GM-based estimations should be considered in the development of diagnostic tools and indicate clinical utility well beyond aiding visual assessments.
Importance: The entry of artificial intelligence into medicine is pending. Several methods have been used for the predictions of structured neuroimaging data, yet nobody compared them in this context.
Objective: Multi-class prediction is key for building computational aid systems for differential diagnosis. We compared support vector machine, random forest, gradient boosting, and deep feed-forward neural networks for the classification of different neurodegenerative syndromes based on structural magnetic resonance imaging.
Design, setting, and participants: Atlas-based volumetry was performed on multi-centric T1-weighted MRI data from 940 subjects, i.e., 124 healthy controls and 816 patients with ten different neurodegenerative diseases, leading to a multi-diagnostic multi-class classification task with eleven different classes.
Interventions: N.A.
Main outcomes and measures: Cohen’s kappa, accuracy, and F1-score to assess model performance.
Results: Overall, the neural network produced both the best performance measures and the most robust results. The smaller classes however were better classified by either the ensemble learning methods or the support vector machine, while performance measures for small classes were comparatively low, as expected. Diseases with regionally specific and pronounced atrophy patterns were generally better classified than diseases with widespread and rather weak atrophy.
Conclusions and relevance: Our study furthermore underlines the necessity of larger data sets but also calls for a careful consideration of different machine learning methods that can handle the type of data and the classification task best.