MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Evaluation of Machine Learning Algorithms Performance for the Classification of Euthymic Pediatric Bipolar Disorder and Healthy Adolescents From Resting-State FMRI Data

T Wang1, Y Guo2, Q Jiao3*, W Cao4, D Cui5, W Gao6, L Su7, G Lu8, (1) ,,,(2) Shandong First Medical University, ,,(3) Shandong First Medical University, Taian, 37, CN, (4) Shandong First Medical University &Shandong Academy of Medical Sciences, Tai'an, 37, CN, (5) Shandong First Medical University, Taian, 37, CN, (6) Zhejiang University School of Medicine, ,,CN, (7) Key Laboratory of Psychiatry and Mental Health of Central South University, ,,CN, (8) Clinical School of Medical College, Nanjing University, ,,CN

Presentations

(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: evaluate temporal measures of functional magnetic resonance imaging (fMRI) as biomarkers to classify euthymic pediatric bipolar disorder(PBD) and healthy controls(HCs) with four machine learning (ML) classifiers.

Methods: was acquired from 20 euthymic PBD patients and 18 age- and gender-matched healthy controls (HCs). Four voxel-based indices, including amplitude of low-frequency fluctuation(ALFF),fractional ALFF(fALFF), Regional Homogeneity(ReHo) and FOur-dimensional Consistency of local neural Activities(FOCA), were calculated and then statistical maps showing significantly differences between the two groups were obtained. The automated anatomical labeling (AAL) atlas template with 116 regions of interest (ROI) was used to extract time series of the values of ALFF, fALFF, ReHo and FOCA from the whole brain maps. Then the extracted ALFF, fALFF, ReHo and FOCA of 116 ROIs were make as features, and input into four classifiers: Random Forest (RF), Support Vector Machine (SVM) , Naïve Bayes (NB) and K-nearest-neighbor (kNN) . The classification accuracy was evaluated with leave-one-out cross-validation (LOOCV) strategy.


Results: classification performance, SVM showed the highest accuracy and AUC in all indices, resulting to be best approach. Moreover, ALFF presented the best discriminating performance among the four indices with the highest accuracy of 84.2% in SVM(KNN:71.0%, NB:34.2%,RF:42.1%) , sensitivity of 72% (KNN:78.9%,NB:64.7%,RF:61.1%), specificity of 95% (KNN:63.1%, NB:66.6%,RF:55%) and an area under curve(AUC) of 0.9 (KNN:0.75, NB:0.74,RF:0.6). The top 5 best discriminating brain regions were vermis, superior cerebellum, supplementary motor area, middle frontal gyrus(MFG), orbital part of MFG.


Conclusion: on fMRI and ML algorithms, the ALFF may be used as potential imaging biomarkers for stable and accurate classification of euthymic PBD patients and healthy adolescents. The results may confirm and establish SVM as a powerful ML tools for the study of PBD-related brain abnormalities. Brain regions presented better classification performance may serve as a priori regions for future analyses.

Funding Support, Disclosures, and Conflict of Interest: This work was supported by the Funds of the National Natural Science Foundation of China (81371531to Qing Jiao; 81901730to Weifang Cao), Key Project of Scientific Research of 12th Five-Year Plan in Medical Research of the Army (BWS11J063 to GL).

Keywords

MRI, Brain, Functional Imaging

Taxonomy

Not Applicable / None Entered.

Contact Email