Parkinson's Disease Diagnosis via Joint Learning From Multiple Modalities and Relations

Abstract

Parkinson's disease (PD) is a neurodegenerative progressive disease that mainly affects the motor systems of patients. To slow this disease deterioration, early and accurate diagnosis of PD is an effective way, which alleviates mental and physical sufferings by clinical intervention. In this paper, we propose a joint regression and classification framework for PD diagnosis via magnetic resonance and diffusion tensor imaging data. Specifically, we devise a unified multi-task feature selection model to explore multiple relationships among features, samples, and clinical scores. We regress four clinical variables of depression, sleep, olfaction, cognition scores, as well as perform the classification of PD disease from the multi-modal data. The multi-task model explores the relationships at the level of clinical scores, image features, and subjects, to select the most informative and diseased-related features for diagnosis. The proposed method is evaluated on the public Parkinson's progression markers initiative (PPMI) dataset. The extensive experimental results show that the multi-task framework can effectively boost the performance of regression and classification and outperforms other state-of-the-art methods. The computerized predictions of clinical scores and label for PD diagnosis may offer quantitative reference for decision support as well.

Parkinson’s Disease Classification and Clinical Score Regression via United Embedding and Sparse Learning From Longitudinal Data

“…In [31], multimodal data is utilized to improve performance on the classification and regression prediction of Alzheimer's disease via relational regularization and discriminative learning. In [5], the authors perform joint learning from multiple relations and modalities to select the discriminative features for classification and prediction of PD. Multi-cluster feature selection (MCFS) [32] approach first calculates the nearest neighbor graph and then selects the discriminative features that best present the clustering information.…”

“…(3) RSFS [19] (https://github.com/LeiShiCS/RSFS), which concurrently exploits FME and norm to robustly select the discriminative features; (4) the MMSL approach obtained from Lei et al [15], which concurrently conducts classification and regression prediction of PD based on a united multi-task feature selection function that considers the similarity of difference among rows and columns in response matrix; (5) the joint multi-task learning (JMTL) approach from Lei et al [5], which performs classification and prediction of PD based on a united multi-task feature selection function that explores multiple relationships in the response matrix, and (6) the M3T [30] approach based on norm, which learns a feature selection model to gain common relevant features of multiple tasks from every modality; the M3T approach is a particular case of MMSL approach when its two regularization terms set to zero.…”

“…We can also see that the top ten brain regions are not only strongly related to each other but also connected to other brain regions in NC vs. PD and PD vs. SWEDD. However, in [5], most of the top brain regions are only related to each other. Compared with the results in [5], our correlation maps can show richer information and thus are more suitable as a complement for PD diagnosis.…”

“…However, in [5], most of the top brain regions are only related to each other. Compared with the results in [5], our correlation maps can show richer information and thus are more suitable as a complement for PD diagnosis. The possible reason is that the multi-modal data (i.e., GM, L1, and V1) used in this paper is more able to present brain change information more effectively.…”

“…Since multimodal data can offer complementary information for the classification of neurodegenerative diseases, these data have played an increasingly significant role and captured widespread attention [5,6]. For instance, in [7], multimodal data is utilized to raise the classification performance of neurodegenerative disease based on a semi-supervised feature-subject selection approach.…”