Back

Feature-Mask-Based Strategies for Subtype-Specific Freezing of Gait Detection using CNNs

Yu, X.; Martens, K. E.; Arami, A.

2025-10-04 bioengineering
10.1101/2025.05.26.656145 bioRxiv
Show abstract

ObjectiveFreezing of gait (FOG), a disabling symptom of Parkinsons disease, varies in manifestations and motion contexts. Its heterogeneity motivates subtype categorization such as manifestation-specific subtypes (akinesia, trembling or shuffling) or motion-specific subtypes (gait-initiation, walking or turning). Despite numerous promising deep learning FOG detection studies, few consider FOG heterogeneity. It remains unclear whether different subtypes require distinct detection strategies, and whether tailoring subtype-specific models could enhance detection generalizability across subtypes. Methods: To address these questions, we categorize FOG data into manifestation- or motion-specific subtypes and derive their corresponding detection strategies as interpretable feature masks. We then propose a feature-mask-based CNN that explicitly embeds the identified strategies. Using waist-mounted 3D accelerometer data, a general CNN and subtype-specific CNNs are trained. Results: According to feature-mask analysis, motion-specific subtypes share a common detection strategy, whereas manifestation-specific subtypes require distinct strategies. Manifestation models exhibit enhanced generalizability across subtypes compared to the general model, boosting the overall average FOG detection sensitivity by 24.95%{+/-}9.80% and specificity by 18.29%{+/-}8.71%. Conversely, motion models reduce the overall FOG sensitivity by 1.89%{+/-}8.74% and specificity by 5.17%{+/-}10.76%. Conclusions: The detection strategy is mainly driven by manifestation composition of the data. The general model favors the dominant manifestation-specific subtype group(s), a bias corrected by tailored manifestation-specific strategies. No comparable benefit arises from motion models due to their similar manifestation compositions. Significance: This study interpretably reveals the detection strategies required by different FOG subtypes and demonstrates the effectiveness of subtype-specific tailoring in improving FOG detection generalizability.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.