The morphology of thalamic subnuclei in Parkinson's disease and the effects of machine learning on disease diagnosis and clinical evaluation.

In Parkinson's disease (PD), the thalamus plays an important role in pathogenesis and disease symptoms; however, the morphological changes in thalamic subnuclei have not been clearly investigated. And there are still many challenges in individual PD diagnosis, especially clinical condition evaluations. Structural magnetic resonance imaging (MRI) was performed on 131 PD patients and 69 healthy controls (HC), and the volumes of 25 thalamic subnuclei were evaluated by FreeSurfer and a newly developed thalamus segment algorithm. Then, the individual PD diagnosis and clinical condition prediction were conducted on support vector machines (SVM) classification or regression. The bilateral thalami were enlarged; the volumes of 21 of 25 left thalamic subnuclei and 20 of 25 right thalamic subnuclei were increased, accompanied by 2 left nuclei atrophy. An accuracy of 95% with sensitivity of 97.44%, and specificity of 90.48% was achieved in PD diagnosis. United Parkinson's disease Rating Scale (UPDRS) III, limb bradykinesia, and axial akinetic symptoms score prediction were obtained with Pearson correlation coefficient of 0.5497, 0.5382, and 0.5911, respectively; however, the results of tremor, rigidity, and speech prediction were limited. Finally, accuracies of 76.92% were achieved in the UPDRS III improvement prediction. These findings confirmed that numerous left and right thalamic subnuclei were enlarged, accompanied by a few atrophies. The individual PD diagnosis, symptom, and clinical improvement prediction could be achieved based on morphology of thalamic subnuclei via machine learning.