ABSTRACT
The detection of structural changes in the brain by employing SPECT with 123I-Ioflupane images can help in the early detection of Parkinson disease. Clinical data analysis has made significant achievements in disease identification, diagnosis, and prevention. Nevertheless, such clinical data generally suffers from highly imbalanced classes, with one class consisting of a large number of samples while the others consist of only a few. Imbalanced data classification is one of the most challenging problems in machine learning. In this chapter, an approach based on the synthetic minority over-sampling technique (SMOTE), to solve the imbalanced data problem, and ensemble machine learning methods are utilized to improve the prediction rate of early Parkinson disease and the detection rate of Scan without Evidence of Dopaminergic Deficit (SWEDD) subjects. Experimental results revealed that this technique could significantly improve the specificity rate while preserving high sensitivity and overall accuracy rates. Furthermore, a rotation forest with a PCA-based AdaBoost (RFBoostRotF) improves overall performance when compared to traditional machine learning methods. The proposed model achieved a 0.994 F-measure, 0.998 AUC value, 99.75% sensitivity, 99.03% specificity, and 99.41% overall accuracy in detection of early PD; and a 0.994 F-measure, 0.999 AUC value, 99.87% sensitivity, 98.95% specificity, and 99.42% overall accuracy in detection of SWEDD subjects, distinguishing them from Parkinson disease subjects. Therefore, it can be seen that the proposed method can help the clinician in making a correct diagnosis of early PD.
