Machine Learning Based Classification Using Clinical and DaTSCAN SPECT Imaging Features: A Study on Parkinson's Disease and SWEDD

• Published in: IEEE transactions on radiation and plasma medical sciences Vol. 3; no. 2; pp. 170 - 177
• Main Authors: Mabrouk, Rostom, Chikhaoui, Belkacem, Bentabet, Layachi
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.03.2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: <italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">K -nearest neighbor (<italic xmlns:ali="http://www.niso.org/schemas/ali/1.0/" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance">K -NN); Artificial intelligence; Basal ganglia; Biomedical imaging; Brain diseases; Central nervous system diseases; Classification; Computed tomography; Diseases; Dopamine receptors; Emission analysis; Image classification; Learning algorithms; logistic regression (LR); Machine learning; machine learning (ML); Medical imaging; Model accuracy; Movement disorders; multilayer perceptron (MLP); Neostriatum; Neurodegenerative diseases; Parkinson's disease; Parkinson’s disease (PD); Photon emission; Radio frequency; random forest (RF); scan without evidence of dopaminergic deficit (SWEDD); Single photon emission computed tomography; single-photon emission computed tomography (SPECT); support vector machine (SVM); Support vector machines; Tomography
• ISSN: 2469-7311; 2469-7303
• DOI: 10.1109/TRPMS.2018.2877754

DatSCAN single-photon emission computed tomography (SPECT) imaging is a reliable method to assess dopaminergic transporter in degenerative Parkinsonism. Scan without evidence of dopaminergic deficit (SWEDD) subjects are clinically diagnosed as Parkinson's disease (PD) patients although the SPECT imaging does not show any negro-striatal abnormality. In this paper, five models of machine learning were used to carry out binary classification [healthy control (HC)/PD] using clinical assessment and image-derived features applied thereafter on SWEDD group as a potential application of motor and nonmotors features in understanding PD characteristic in this group. The nested cross-validation was an essential component to select reliable models. A high accuracy was achieved for the five models (75.4%-78.4% for motor features and 71%-82.2% for nonmotor features) in binary classification (HCs versus PD). Accordingly, we demonstrate the suitability and usefulness of ML models to carry out binary classification of SPECT data. Cross all models applied on SWEDD group, 17.6% of patients were classified as PD motor disorder lookalikes, 27.4% were classified as having a beginning nonmotor abnormality of PD, and 3.9% were classified as having both motor and nonmotor PD features. However, the interpretability of SWEDD predicted condition must be carefully considered.