Automated segmentation of multiple sclerosis lesions, paramagnetic rims, and central vein sign on MRI provides reliable diagnostic biomarkers
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Multiple sclerosis (MS) is characterized by central nervous system lesions detectable via MRI. Existing diagnostic criteria incorporate presence of white matter lesions, but specificity can be improved using MS-specific imaging biomarkers, including paramagnetic rim lesions (PRLs) and central vein sign (CVS). However, manual segmentation of lesions, PRLs, and CVS is time-consuming and subjective. We propose a fully-automated joint segmentation method called Automated Lesion, PRL, and CVS Analysis (ALPaCA). We trained ALPaCA using subject-level cross-validation on 47 adults with MS and 50 adults with radiological MS mimics. ALPaCA uses a voxel-wise lesion segmentation method to propose a large set of lesion candidates. Lesion candidates are input into a multi-contrast, multi-label 3D convolutional neural network as 3D patches to produce lesion, PRL, and CVS predictions. When multiple lesions exist within a patch, an attention mechanism identifies which lesion candidate to classify. At the lesion level, ALPaCA achieves cross-validation areas under the receiver operating characteristic curve (AUROCs) of 0.95, 0.91, and 0.87 for lesion, PRL, and CVS classification, outperforming previous methods (all p < 0.001). Correlations between subject-level ALPaCA lesion and PRL scores with manual counts are higher than those of previous methods (p < 0.001; p = 0.03). Subject-level ALPaCA PRL and CVS scores are highly associated with MS in logistic regressions, when controlling for age and sex (p < 0.001). ALPaCA allows for fully-automated simultaneous segmentation of MS lesions, PRLs, and CVS using clinically-feasible scans. These segmentations outperform existing methods at the lesion and subject level.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.006 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it