Enterprise AI Analysis

The systematic review of 89 studies revealed several critical insights. Firstly, 52.8% of investigations used proprietary data, limiting reproducibility, though a trend towards open access (20.2%) and hybrid models (19.1%) is emerging post-2018 [24, 25, 26]. Secondly, motor neuron degeneration, particularly ALS (71% of entries), dominates research focus, with imaging prevalent in PSP and biochemical streams in ALS. Proteomics occupies 11% of open repositories but only 2% of closed ones. Thirdly, 128 algorithmic pipelines were extracted, with 43% using ensemble tree frameworks (e.g., gradient boosting with SHAP) achieving median AUROC 0.89 for ALS prognostic tasks, outperforming recurrent networks. Multimodal integration improved balanced accuracy by 9.4% [25, 32, 71]. Key attributes included cortical thickness, diffusion fractional anisotropy, multiscale entropy of gait, neurofilament concentrations, and polygenic risk vectors. Fourthly, interpretability techniques were present in 45% of open projects versus 17% of closed ones, with SHAP dominating. Finally, the median reported AUROC was 0.84 (IQR 0.75–0.91), with open status showing a slightly higher median (0.86 vs 0.83), though not statistically significant (p = 0.18). This suggests that while openness aids external validation, it doesn't solely guarantee superior performance. (From Sections 4.1-4.5)

The review identifies several crucial future directions to advance MND research. There is a need for robust data harmonization across multi-center repositories, enforcing standardized consent, privacy-preserving record linkage, and versioned schemas. The development of purpose-built graph neural networks (GNNs) is essential for cross-modal biomarker fusion, encoding pathway priors, and integrating diverse omics (transcriptomic, proteomic, epigenetic) with imaging metrics. Rigorous multi-site validation with pre-registration, time-aware splits for prognosis, and heterogeneity stress tests across scanners/cohorts are critical for ensuring reproducibility and generalizability. Interpretability, particularly through XAI frameworks like SHAP, must be a core component of reporting, mapping influential features to biological pathways with expert review. Ultimately, these steps aim to translate promising metrics into validated evidence suitable for clinical translation, fostering a shift towards integrative, transparent, and reproducible MND research with a focus on mechanistic relevance and precise patient stratification. (From Section 5 and Proposed steps in Table 1)