The Science Behind NeuroHawk

A transparent look into our data-driven, quantifiable approach to ALS risk assessment.

Methodology

NeuroHawk’s predictive power comes from a logistic regression model trained on a 500-sample simulated neuroimaging dataset. This dataset was carefully designed to reflect real-world clinical variations.

The model analyzes three key metrics derived from diffusion MRI, which are known to be associated with neurodegeneration:

Cervical Spine Cross-Sectional Area (CSA_mean): A measure of spinal cord atrophy.
Fractional Anisotropy (FA): Indicates the directional coherence of water diffusion, reflecting white matter integrity.
Mean Diffusivity (MD): Measures the average magnitude of water diffusion, which can increase with tissue damage.

Since these metrics are automatically extracted from routine tract-based MRI post-processing, NeuroHawk is designed to be fully implementable in existing radiology workflows without requiring additional, complex procedures.

Model Coefficients

The final optimized model provides a clear and clinically intuitive framework where reduced CSA/FA and elevated MD strongly increase the predicted likelihood of ALS.

Parameter	Coefficient Value	Clinical Implication
Intercept (b₀)	+6.18	Baseline model bias
CSA_mean (b₁)	–0.175	↓ CSA increases ALS probability
FA (b₂)	–4.49	↓ FA increases ALS probability
MD (b₃)	+5.60	↑ MD increases ALS probability

Clinical Decision Support

To provide a high-specificity tool for clinicians, we established a clear threshold and reference values for identifying high-risk patients.

Probability Threshold

A probability threshold was set at p = 0.90 (corresponding to a logit cut-point of 2.197). A prediction above this value classifies a patient as “ALS-Present,” indicating a high-confidence signal that warrants further investigation.

90th-Percentile Clinical Reference Values

These values represent high-specificity biomarkers that a clinician can directly compare against their patient’s diffusion MRI metrics to quickly contextualize results.

CSA Mean

61.38 mm²

Fractional Anisotropy

0.56

Mean Diffusivity

0.97×10⁻³ mm²/s

A Step Towards Evidence-Based Medicine

These results support NeuroHawk as a rapid, quantifiable, and imaging-driven decision support tool that translates advanced diffusion metrics into actionable ALS risk predictions, empowering clinicians with data-driven insights.