AI and Neurological Rare & Neuroimmune Diseases ── Parkinson's ── 2026-11-13
Parkinson's disease and parkinsonism: treatment, trials and AI
- Abstract
Introduction: Multimodal deep learning has been proposed as a low-cost method for early screening of Parkinson's disease (PD), with most recent frameworks reporting accuracies exceeding 96% using k-fold cross-validation instead of a single fixed train/test partition.
Methods: We introduce ParkFuse, a trust-gated late-fusion framework that assigns each modality a matched representation-classical learned acoustic descriptors for voice, geometric stroke and tremor descriptors for handwriting, and residual convolutional networks for images-combined through soft-voting late fusion. Out-of-fold predictions are obtained via stratified, grouped k-fold cross-validation, and a quantitative shortcut audit tests whether classes are separable by trivial acquisition properties alone, yielding a trust gate that excludes leakage-suspect modalities from fusion.
Results: On a public 10-modality PD dataset, the audit reveals that imaging modalities achieve a near-optimal train-validation gap, whereas the electrocardiogram (ECG) folder contains files that are byte-identical across both classes-a signature of leakage rather than disease. In contrast, sustained-vowel voice recordings remain the strongest valid predictor (achieving out-of-fold AUCs of 0.98 and 0.96). A controlled ablation study shows that modality-matched features outperform a uniform image-style network on voice and drawing by as much as 0.43 in AUC. Discussion We argue that headline multimodal PD accuracies must always be reported alongside leakage diagnostics and that voice remains the most trustworthy modality for screening.
Journal IF-equivalent: 3.4 (OpenAlex 2-year mean citedness, value as of 2026-11-12, retrieved 2026-11-13; not the official Clarivate IF)Reference: Oli NM, Rahman A, Aziz AKMRB, Dutta S, Kader A, Sharfuddin M, et al. ParkFuse: a trust-gated late-fusion framework with shortcut auditing for multimodal Parkinson’s disease detection. Front Med. 2026 Nov 11;13:1905869. doi:10.3389/fmed.2026.1905869.Checked: Abstract only - Abstract
BACKGROUND: Early and accurate diagnosis of Parkinson's disease (PD) remains challenging due to clinical overlap with other parkinsonian syndromes. DaTscan single-photon emission computed tomography (SPECT) is widely used during diagnosis; however, a subset of clinically diagnosed patients presents scans without evidence of dopaminergic deficit (SWEDD), creating diagnostic ambiguity. This study developed and validated a deep learning (DL)-based multiclass framework to distinguish PD, SWEDD, and normal controls using DaTscan SPECT images. METHODS: DaTscan images from the Parkinson's Progression Markers Initiative (PPMI) were used for model development, with external validation performed using an independent cohort from Chiayi Christian Hospital. AlexNet, EfficientNet-B0, and Vision Transformer (ViT) architectures were trained and evaluated using accuracy, precision, recall, and F1-score. The best-performing model was integrated into an automated diagnostic tool, the Parkinson's Disease Detector App. RESULTS: EfficientNet-B0 achieved the highest accuracy for three-class classification on the PPMI dataset (94.2%), outperforming ViT (92.0%) and AlexNet (89.5%). External validation demonstrated strong generalizability (accuracy 92.4%, F1-score 95.8%). The diagnostic app achieved 90.6% accuracy in distinguishing PD and SWEDD from controls. CONCLUSION: The proposed DL framework enables accurate multiclass classification of DaTscan SPECT images and shows promise as an objective adjunct for PD diagnosis. Multicenter validation is warranted.
Journal IF-equivalent: 2.0 (OpenAlex 2-year mean citedness, value as of 2026-11-12, retrieved 2026-11-13; not the official Clarivate IF)Reference: Kenneth MJ, Hsu YL, Cheng CY, Mukundan A, Hsieh MC, Lai LM, et al. Enhancing diagnosis of Parkinson’s Disease through a deep learning–based multiclass interpretation of DaTscan SPECT images. Psychiatry Res Neuroimaging. 2027 Mar;365:112385. doi:10.1016/j.pscychresns.2026.112385. PMID: 43014087.Checked: Abstract only