Optimized Vmd-1d CNN Framework for Real-Time and Early Detection of Parkinson’s Disease from Gait Dynamics

Dr. Chejarla Raja Shekhar¹, Gormi Pulicherla², Poondla Baby Priya³, BalaGanesh Vullikanti⁴, A P Kamalesh⁵
¹Associate Professor, Dept of Information Technology,SV College Of Engineering, Tirupati, India.
²B.Tech, Dept of Information Technology, SV College Of Engineering, Tirupati, India.
³B.Tech, Dept of Information Technology, SV College Of Engineering, Tirupati, India.
⁴B.Tech, Dept of Information Technology, SV College Of Engineering, Tirupati, India.
⁵B.Tech, Dept of Information Technology, SV College Of Engineering, Tirupati, India.
Email:¹rajasekhar.ch@svce.edu.in, ²gormipulicherla@gmail.com , ³poondlapriya426@gmail.com ,
⁴balaganeshvullikanti@gmail.com , ⁵kamaleshap3@gmail.com
Corresponding Author:-Dr. Chejarla Raja Shekhar

Abstract-Parkinson's disease (PD) significantly impairs gait dynamics, making accurate diagnosis challenging through traditional clinical assessments like the Unified Parkinson's Disease Rating Scale (UPDRS), which suffer from subjectivity and time complexity. Existing systems utilize decomposition techniques— Empirical Mode Decomposition (EMD), Empirical Wavelet Transform (EWT), and Variational Mode Decomposition (VMD)—applied to PhysioNet gait signals from vertical ground reaction force (VGRF) sensors, followed by statisticalfeature extraction (mean, min, max, skewness, kurtosis) and classification via machine learning (SVM, DT,KNN, ANN) and deep learning (LSTM, BiLSTM, 1D CNN) models. The VMD-1D-CNN combination achieves good accuracy, sensitivity, and specificity. However, limitations include EMD's modal mixing and noise sensitivity, EWT's boundary detection issues leading to slower processing, VMD-1D-CNN's highcomputational cost and memory usage unsuitable for real-time clinical deployment, and lack of focus on early PD detection.The proposed system addresses these by optimizing VMD hyperparameters (α=2000, K=3), integrating advanced feature refinement, and deployinglightweight 1D-CNN architectures with efficient optimizers (Adam, ReLU) for real-time processing. Benefits encompass enhanced early PD detection,reduced computational complexity for wearable  deployment, improved robustness across diverse gait datasets, and superior generalizability, enabling precise,non-invasive clinical screening with minimal latency.Keywords: Parkinson's disease, generalizability,EmpiricalMode Decomposition (EMD),Empirical Wavelet Transform (EWT), Variational Mode Decomposition (VMD),PhysioNet gaitsignals.