Using deep-learning to predict outcome of patients with Parkinson’s disease K. H. Leung, M. R. Salmanpour, A. Saberi, I. S. Klyuzhin, V. Sossi, A. K. Jha, M. G. Pomper, Y. Du, and A. Rahmim Abstract– There are currently no established disease modifying therapies for PD, and prediction of outcome in PD to power clinical studies is a very important area of research. Assessment of PD is informed by imaging the dopamine system with dopamine transporter (DAT) single-photon emission computed tomography (SPECT) imaging and by the presence of key symptoms. Recently, deep-learning based methods have shown promise for medical image analysis tasks and disease detection. The purpose of this study was to develop a deep- learning based approach to predict outcome of patients with PD using longitudinal clinical data containing imaging and non- imaging information. Features were first extracted from the clinical data by the proposed deep-learning based approach and then combined to predict motor performance (MDS-UPDRS-III) in year 4. The performance of the proposed approach was evaluated via a 10-fold cross-validation. We evaluated the performance of the network on the basis of mean absolute error (MAE) between the predicted and true MDS-UPDRS part III scores in year 4. The proposed approach yielded a MAE of 4.33±3.36 when given only imaging features, 3.71±2.91 when given only non-imaging features, and 3.22±2.71 when given all input data. While the approach given only non-imaging input data outperformed the approach given only imaging data, we found that the performance of the proposed approach substantially improved when given both imaging and non- imaging information. Our results indicate that the addition of imaging data to non-imaging clinical data is helpful for the prediction of outcome in patients with PD. The proposed approach that incorporated both imaging and non-imaging clinical data shows significant promise for prediction of outcome in patients with PD. I. INTRODUCTION Parkinson’s disease (PD) is the second most common neurodegenerative disorder which is characterized by neuronal loss of dopaminergic neurons in the substantia nigra [1]. PD is a progressive movement disorder where the loss of dopamine levels can cause progressive motor and non-motor symptoms. Patients with PD may exhibit motor symptoms, such as resting tremor, bradykinesia, muscle stiffness and postural instability, as well as non-motor symptoms including cognitive problems and autonomic nervous system dysfunction usually occurring in the later stages of the disease [1]. The diagnosis of PD is Manuscript received December 19, 2018. K. H. Leung, M. G. Pomper, and Y. Du are with Johns Hopkins University, Baltimore, MD, USA; M. R. Salmanpour is with Amirkabir University of Technology, Tehran, Iran; A. Saberi is with Islamic Azad University, Tehran, Iran; I. S. Klyuzhin and V. Sossi are with the University of British Columbia, Vancouver, BC, Canada; A. K. Jha is with Washington University in St. Louis, St. Louis, MO, USA; A. Rahmim was with Johns Hopkins University, Baltimore, MD, USA. He is now with the University of British Columbia, Vancouver, BC, Canada. informed by the presence of such key symptoms and by imaging the dopamine system with 123I-isoflupane-dopamine transporter (DAT) single-photon emission computed tomography (SPECT) [1]. There are currently no established disease modifying therapies for PD, and prediction of outcome in PD to power clinical studies is an active area of research [1]-[5]. Due to the need for identifying biomarkers of PD progression, the Parkinson’s Progression Markers Initiative (PPMI) has made available longitudinal clinical data of patients with PD that included a database of non-imaging clinical measures of PD and DAT-SPECT images [6]. Recently, deep-learning based methods have shown promise for medical image analysis tasks and disease detection [7]. Therefore, in this project, our aim is to develop a deep- learning approach to predict motor outcome of patients with PD by incorporating both imaging and non-imaging information. We aim to develop this deep-learning approach as a prognostic tool that may further characterize patients into different groups. This could lead to determining different treatments or therapy regimens for each patient to ultimately reduce symptoms and to delay the disease progression. II. MATERIALS AND METHODS A. Patient data The longitudinal clinical data, including DAT-SPECT images and clinical measures of patients with PD, were extracted from 198 patients (144 males and 54 females, mean age 67.6±9.98 years, range [39,91]) in the PPMI database. DAT-SPECT images and clinical measures from year 0 (baseline) and year 1 were used as predictors. The non- imaging clinical measures included movement disorder society unified Parkinson’s disease rating scale (MDS- UPDRS) – part III from both year 0 and year 1 as well as age, gender, and diagnosis duration with respect to time of diagnosis and time of appearance of symptoms. For the prediction task, we define the composite MDS-UPDRS-III score in year 4 as outcome. The DAT-SPECT images were preprocessed by selecting a continuous segment of 21 image slices of each image where the center slice had the highest relative intensity in the trans- axial direction. The images were then zero padded resulting in 128 x 128 x 21 sized images for both year 0 and year 1. B. Varying the input to the proposed approach Given the availability of a heterogenous longitudinal dataset, we developed several deep-learning based approaches that used different input data. The first method uses only information from DAT-SPECT images. The second method