In a recent article published in the magazine Scientific ReportsResearchers proposed an approach based on quantum machine learning (QML) and deep ensemble learning for Alzheimer's disease detection.
background
Alzheimer's disease is one of the most chronic neurodegenerative diseases threatening public health worldwide. This degenerative brain disorder impairs important mental abilities, including memory. Patients with Alzheimer's disease face a variety of challenges, including motor dysfunction, memory loss, behavioral disorders, and cognitive impairment.
Therefore, early diagnosis of AD is essential for timely intervention and medication to improve patients' quality of life.Structural magnetic resonance imaging (MRI), a subclinical diagnostic method, has been recognized as a common imaging biomarker to classify and identify the stage of AD into non-dementia, very mild dementia, moderate dementia, and mild dementia.
Deep Learning (DL) techniques have shown significant improvements in disease detection, classification, and detection tasks. Ensemble learning can further enhance the performance of these DL models. Moreover, QML, the intersection of quantum computing and classical ML, has also attracted significant attention from the scientific community due to its promising speed, scalability, expressiveness, and flexibility.
Quantum computing offers more efficient models for various disease classification tasks compared to classical ML approaches. Many studies have demonstrated the superiority of QML algorithms over classical ML algorithms in several applications such as healthcare. However, the full potential of quantum computing has not been applied to the task of Alzheimer's disease classification.
the study
In this study, the researchers proposed an ensemble DL model based on QML classifiers for classifying Alzheimer's disease. The proposed model can detect the stage of Alzheimer's disease from brain MRI by using an ensemble of customized versions of VGG16-ResNet50 DL models as feature extraction and then applying a QML algorithm for classification.
The aim of this study is to integrate various source data to develop an efficient model for better outcome prediction to address the challenge of Alzheimer's disease classification. We combined the MRI datasets from the Alzheimer's Disease Neuroimaging Initiative II (ADNI2) and the Alzheimer's Disease Neuroimaging Initiative I (ADNI1) obtained from Kaggle to perform Alzheimer's disease classification.
Using a customized version of the Ensemble DL algorithm/VGG16 and ResNet50 models, key Alzheimer's disease features extracted from the merged images/ADNI MRI imaging data were first combined and fed into a QML classifier.
The QML classifier/Quantum Support Vector Machine (QSVM) classified the patients into very mild dementia, moderate dementia, mild dementia, and non-dementia.Thus, the proposed approach allows for decisions to be made in a more diverse, thorough, and reliable manner.
The researchers evaluated the performance of the proposed model using multiple metrics, including recall, precision, F1 score, area under the curve, and accuracy, and investigated the effectiveness of QML in effectively improving the computational efficiency of the model. They also investigated the effectiveness of the model against other techniques, such as Convolutional Neural Networks (CNNs), Inception-ResNet-V2, VGG16, AlexNet, Random Forest, and Adaptive Boosting Classifiers.
Additionally, the researchers evaluated the performance of an ensemble learning method using QSVM and classical SVM classifiers. They used a 5-qubit quantum simulator/hardware and the QSVM model from the Qiskit library. The QSVM model was optimized by adding hyperparameters.
The importance of this work
In the results for the individual DL models, ResNet performed poorly based on F1 score and recall, while VGGNet achieved superior results along with the ensemble model. The area under the curve was superior compared to other metrics, with VGGNet, ResNet, and the ensemble model achieving scores of 97.59, 93.70, and 98.84, respectively.
The researchers investigated the effectiveness of SVM-based DL models on the Alzheimer's disease dataset. Results showed that the VGGNet+SVM model achieved an F1 score of 85%, recall of 85.30%, precision of 85.24%, area under the curve score of 89.18%, and accuracy of 85.00%, while the ResNet+SVM model achieved an accuracy of 82.24%, F1 score of 84.73, and recall of 85.43. The proposed ensemble (VGGNet+ResNet) combined with a traditional SVM model achieved the best accuracy of 86.78% and area under the curve score of 90.53%.
The results of DL model with QSVM also showed that the proposed ensemble model + QSVM performed better compared to VGGNet + QSVM and ResNet + QSVM, with the ensemble model + QSVM achieving the highest accuracy of 99.89, precision of 99.25, and area under the curve score of 99.99 on the merged ADNI dataset.
These results validate that the proposed ensemble model of SVM and QSVM outperforms traditional DL with SVM and DL with QSVM. Furthermore, the proposed model achieves the highest accuracy of 99.89 on the combined ADNI1+ADNI2 dataset, surpassing the previous best accuracy of 99.68 achieved by CNN on the OASIS dataset, and outperforming other methods for Alzheimer's disease detection.
In summary, the proposed approach provides an effective solution to support primary care of Alzheimer's disease, especially when MRI scans are blurred and make it difficult for experts to adequately suggest the disease.
Journal Reference
Jenber Belay, A., Walle, YM, Haile, MB (2024). Deep ensemble learning and quantum machine learning approaches for Alzheimer's disease detection. Scientific Reports14(1), 1-10. https://doi.org/10.1038/s41598-024-61452-1, https://www.nature.com/articles/s41598-024-61452-1
