
Multitask Deep Learning for Predicting Parkinson’s Progression and Depression from Multimodal Time Series Data
Author(s) -
Muhammad Junaid,
Mitra Ghergherehchi,
Seunghyun Lee
Publication year - 2025
Publication title -
ieee access
Language(s) - English
Resource type - Magazines
SCImago Journal Rank - 0.587
H-Index - 127
eISSN - 2169-3536
DOI - 10.1109/access.2025.3593254
Subject(s) - aerospace , bioengineering , communication, networking and broadcast technologies , components, circuits, devices and systems , computing and processing , engineered materials, dielectrics and plasmas , engineering profession , fields, waves and electromagnetics , general topics for engineers , geoscience , nuclear engineering , photonics and electrooptics , power, energy and industry applications , robotics and control systems , signal processing and analysis , transportation
The high cases of Parkinson’s disease (PD) in the senior generation require reliable PD progression prediction. PD can be associated with depression as a common non-motor symptom. Depression in Parkinson’s disease (PD) is frequently underdiagnosed, despite its significant impact on patients’ quality of life and caregiver burden. Early detection and identification of predictive factors are essential for timely intervention. This study proposes a novel multitask hybrid deep learning framework that leverages multimodal time-series data—including demographic, motor, and non-motor features—to enhance the detection and understanding of depressive symptoms in PD patients. These three modalities are early fused, and the relevant features of the resulting feature set were extracted using a deep variational autoencoder model. Because MRI data are critical in neurodegenerative disease diagnosis, a 3D CNN module has been used to learn deep features from MRI images, and these deep representations were lately fused with the extracted features of the variational autoencoder. The extracted deep representations were used to build a multitask classifier based on a novel Bidirectional LSTM model. This model has been optimized to predict the three tasks of (1) a four-class NHY scale that defines the severity of Parkinson’s disease, (2) a binary class that determines whether or not a patient is depressed, and (3) a four-class depression scale that indicates the levels of depression. The proposed model has been evaluated on single-task and multitask problems with various time steps. Using a 2-time step in the NHY prediction task, our model obtains the maximum cross-validation and testing accuracies of 96.78% and 94.37%, respectively. The multitask model produced steady increases in testing accuracy of 86.29%, 87.12%, and 95.81% with 2-time step, 4-time step, and 6-time step, respectively. The proposed model investigated 1059 PD patients over six-time steps and 850 MRI images of single-time steps from the Parkinson’s Progression Markers Initiative study (PPMI) dataset.
Accelerating Research
Robert Robinson Avenue,
Oxford Science Park, Oxford
OX4 4GP, United Kingdom
Address
John Eccles HouseRobert Robinson Avenue,
Oxford Science Park, Oxford
OX4 4GP, United Kingdom