Prion diseases are fatal neurodegenerative disorders caused by misfolded prion protein accumulation in the central nervous system. Creutzfeldt-Jakob disease (CJD) is the most common human prion disease and includes sporadic, genetic, and acquired forms; sporadic CJD (sCJD) accounts for the majority of cases. Typical clinical features include rapidly progressive dementia, ataxia, visual disturbances, myoclonus, pyramidal/extrapyramidal signs, and, in late stages, akinetic mutism. Antemortem diagnosis remains challenging despite contributions from MRI and cerebrospinal fluid assays, and accurate early diagnosis has implications for infection control, exclusion of treatable mimics, and end-of-life planning.
Electroencephalography (EEG) offers a low-cost, low-burden biomarker modality and has been applied in diagnostic classification of multiple neurological disorders. This study investigates whether resting-state scalp-recorded EEG-derived indices can support classification of sCJD versus Alzheimer’s disease (AD) and healthy older adults.
The primary aim was to classify 6 patients with sporadic Creutzfeldt-Jakob disease from 23 patients with Alzheimer’s disease and 10 healthy older adults using resting-state EEG-derived features. The authors sought to evaluate multiple EEG indices, apply feature selection, and assess classifier performance with validation and robustness analyses.
The cohort comprised 6 sCJD patients, 23 AD patients, and 10 healthy older adults. From resting-state scalp EEG recordings the investigators derived three classes of indices:
These indices provided candidate features for machine learning classification.
Prior to graph-metric calculation, the study standardized power spectrum and SL values for each subject and each frequency band. An exponential transformation was applied to standardized values. Graph metrics were then computed from transformed SL values, yielding a feature set composed of spectral, connectivity, and network characteristics across frequency bands.
All processed EEG features used in the machine learning analyses are provided as supporting information. Raw EEG signals are not publicly available due to ethical and privacy restrictions; requests for access to raw data must be submitted to the Institutional Review Board of the National Center of Neurology and Psychiatry and will be reviewed according to institutional and legal requirements.
Recursive Feature Elimination (RFE) was used to select features for classifier training. The authors report that a 12-dimensional feature subset produced the highest classification accuracy in their primary analyses. Details on the classification algorithm family and hyperparameters are not reported in the abstract and introduction; those specifics are contained in the full article.
Multiple validation strategies were applied to assess classifier reliability:
Collectively, these validation steps support that the reported classification performance was not solely the result of overfitting to a single split, although performance decreased under stricter leakage control.
Key limitations reported by the authors include the small number of sCJD cases (n = 6) and the overall sample size, which constrain generalizability. The authors explicitly state that further validation in larger independent cohorts is required to establish clinical reliability and broader applicability of the proposed EEG-based classifier.
Regarding data availability, processed feature matrices used for machine learning are available as supporting information. Raw EEG signals cannot be shared publicly because participants did not authorize such release; investigators seeking raw data should contact the Institutional Review Board of the National Center of Neurology and Psychiatry.
Resting-state scalp-recorded EEG-derived indices—including spectral power, Synchronization Likelihood, and graph-theory metrics—provided candidate features that could distinguish sCJD from AD and healthy aging in this small cohort. Performance reached high accuracy in primary analyses and remained meaningful under stricter nested cross-validation and robustness checks, but the authors emphasize that larger, independent cohort validation is necessary before clinical application. Given EEG’s accessibility and low burden, these findings motivate further study of EEG-based biomarkers for antemortem classification of prion disease.