Objective and quantified instrumented measurement of ataxia in children with Friedreich ataxia

measurement of ataxia in children with Friedreich ataxia

0:00

We're excited to share insights into our latest research, which is set to significantly enhance how we assess Friedreich’s Ataxia (FRDA) in children. This work introduces a novel multi-device correction framework designed to provide clinicians with a more precise tool for monitoring disease progression and evaluating therapeutic interventions in young patients.

The Challenge: Distinguishing FRDA from Normal Development FRDA is a progressive neurodegenerative disorder affecting motor coordination, balance, and speech. Accurate assessment of its severity is crucial for effective management. While the modified Friedreich’s Ataxia Rating Scale (mFARS) is the standard clinical tool, its scores in pediatric patients are often complicated by developmental ataxia (DA). DA refers to the normal motor variability that occurs as children grow, making it challenging to differentiate the pure impact of FRDA from typical developmental movements. This can lead to inaccuracies in severity assessments.

Our Solution: Advanced AIMs Scores and a Novel Correction Framework To overcome this, we've built upon our prior work with Ataxia Instrumented Measures (AIMs) scores, which provide objective, data-driven severity assessments using specialized inertial measurement units (IMUs). Our new study extends the utility of AIMs scores by introducing a sophisticated correction framework to isolate pure FRDA severity from developmental confounders.

We utilize three specialized IMU devices, each tailored for specific motor domains and equipped with unique machine learning algorithms for correction:

• AIM-C (Cup): Designed for upper-limb assessment during a simulated drinking task. For AIM-C, we use a reinforcement learning (RL)-based technique to optimize filter parameters and minimize DA effects. This approach aligns features from child patient signals with predefined ranges derived from adult signals, assuming DA effects are absent in adults. The RL agent adjusts parameters for median and wavelet denoising filters to achieve this alignment.

• AIM-S (Spoon): Used for fine motor skill evaluation via a self-feeding task. For AIM-S, a Bayesian optimization framework is employed. This method fine-tunes filtering processes by minimizing a loss function that quantifies the deviation of corrected child signals from predefined feature ranges based on adult data.

• AIM-P (Pendant): Designed to evaluate balance and gait during static and dynamic tasks like the Romberg’s test. For AIM-P, a multi-layer perceptron (MLP) framework is used to disentangle the DA component from the observed signals, aiming to isolate the pure FRDA component. This framework considers the child's signal as a combination of control, DA, and pure FRDA components.

  
  

These tailored methodologies ensure a robust correction process, aligned with the unique characteristics of each dataset.

Key Findings: More Accurate Assessments for Children Our results demonstrate that this multi-device correction framework effectively removes the developmental factor from severity scores, providing refined scores that represent the pure FRDA effect.

• We utilized a proportional correction metric, which is the ratio of the severity difference (original mFARS vs. corrected AIMs score) to the original mFARS score.

• A key finding was the negative correlation between proportional correction and age. This confirms our hypothesis that younger patients exhibit higher proportional corrections due to more pronounced DA effects, which diminish as children grow older. This trend was observed consistently across all three devices and in child controls.

• AIM-C exhibited the strongest correlation (Pearson r = -0.69, Spearman ρ = -0.54) among FA patients, suggesting its correction framework is particularly robust. AIM-S and AIM-P also showed meaningful, consistently negative correlations.

  
  

Looking Ahead This work highlights the potential of leveraging advanced signal processing and machine learning techniques to enhance the clinical utility of severity assessments in pediatric FRDA patients. By isolating DA effects, the corrected AIMs scores offer clinicians a robust and accurate tool for monitoring disease progression, evaluating therapeutic outcomes, and guiding clinical decision-making.

Future research will expand validation across larger patient cohorts and explore the potential for a unified model integrating data from all three devices. This is a significant step forward in developing objective, precise, and clinically meaningful tools for pediatric neurodegenerative disorder assessments.