Automated Parkinson’s disease identification from MRI images using InceptionV3 transfer learning and ensemble stacking

Parkinson’s disease (PD) is a neurological disorder caused by a lack of dopamine. It affects the connections between the body and brain, damaging motor functions. The main symptoms are rigidity and tremors. According to WHO reports, 8.5 million individuals were affected by this disease in 2019. Some research estimates indicate that 25.2 million individuals will suffer from it by 2050. Most people affected by this disease are 60 years old, where PD is an aging disease. Disease detection is very difficult, as it has symptoms similar to those of old age (above 50 years). Therefore, early disease identification is essential for improving patient outcomes and ensuring appropriate treatment. Imaging tools and techniques play a vital role in the early identification of neurological diseases such as PD and AD. Advanced machine learning (ML) and Deep Learning (DL) models have been useful for disease identification using imaging technologies in recent years. In this study, we analyzed and extracted features from PPMI PD MRI scans using the Inception V3 (IV3) Transfer Learner (TL) model and stacking-based (Ensemble Learner) classifiers. ML models such as SVM (RBF), MLP (50,50), MLP (70,70), and stacking are used IV3 PD MRI scan features and provide the performance results. A 5-fold cross-validation is a crucial implementation for dataset accuracy stability. The experimental results show that the stacking classifier achieved the best performance, with an AUC of 0.999, classification 98.4% -Accuracy (CA), 98.4%-F1-Score, and 98.7%-recall and 98.3%-Presession. The proposed model shows high accuracy and reliable prediction of PD from MRI images. The proposed model efficiently differentiates PD patients from healthy CN subjects. This was verified by comparing the confusion matrix and ROC curve. Future studies can employ larger multihospital MRI databases to increase the reliability and generalizability of the proposed system. The utilization of DL models along with multimodal clinical data can be suggested to increase the accuracy of predicting the early stages of PD.