Algorithm 2 PD Detection using DL

1:Input: Patient data (voice recordings), Trained DL model 2:Output: Predicted class label (PD/Healthy), Confidence score 3:procedure Preprocessing 4:    Format raw audio data (e.g., WAV format at 44.1 kHz) 5:    Remove background noise using spectral gating or adaptive filtering 6:    Normalize audio amplitude to range $[-1,1]$ 7:    Segment speech into frames and extract relevant features (e.g., MFCC, pitch, jitter) 8:end procedure 9:procedure Disease Detection 10:    Perform 10-fold cross-validation to mitigate class imbalance 11:    Split dataset into training and test sets 12:    Train DL model (e.g., CNN, LSTM, Transformer) on extracted features 13:    Predict labels: PD or HC 14:end procedure 15:procedure Evaluation 16:    Evaluate using metrics: F1-score, accuracy, sensitivity, specificity 17:    Compare performance across different datasets (e.g., multiple voice recordings) 18:end procedure