Accurately recognising patients with progressive mild cognitive impairment (pMCI) who will develop Alzheimer’s disease (AD) in subsequent years is very important, as early identification of those patients will enable interventions to potentially reduce the number of those transitioning from MCI to AD. Most studies in this area have concentrated on high-dimensional neuroimaging data with supervised binary/multi-class classification algorithms. However, neuroimaging data is more costly to obtain than non-imaging, and healthcare datasets are normally imbalanced which may reduce classification performance and reliability. To address these challenges, we proposed a new strategy that employs unsupervised novelty detection (ND) techniques to predict pMCI from the AD neuroimaging initiative non-imaging data. ND algorithms, including the k-nearest neighbours (kNN), k-means, Gaussian mixture model (GMM), isolation forest (IF) and extreme learning machine (ELM), were employed and compared with supervised binary support vector machine (SVM) and random forest (RF). We introduced optimisation with nested cross-validation and focused on maximising the adjusted F measure to ensure maximum generalisation of the proposed system by minimising false negative rates. Our extensive experimental results show that ND algorithms (0.727 ± 0.029 kNN, 0.7179 ± 0.0523 GMM, 0.7276 ± 0.0281 ELM) obtained comparable performance to supervised binary SVM (0.7359 ± 0.0451) with 20% stable MCI misclassification tolerance and were significantly better than RF (0.4771 ± 0.0167). Moreover, we found that the non-invasive, readily obtainable, and cost-effective cognitive and functional assessment was the most efficient predictor for predicting the pMCI within 2 years with ND techniques. Importantly, we presented an accessible and cost-effective approach to pMCI prediction, which does not require labelled data.

Novelty detection (ND) has gained attention in many applications for its effectiveness in dealing with imbalanced data. Many ND algorithms have been proposed. For example, the level set boundary description (LSBD) algorithm can accurately estimate a boundary around normal data which is subsequently used to detect novelties. However, the computational complexity and the convergence time of the LSBD algorithms increases substantially when data dimensionality increases. To solve those challenges, we propose an Integrated Autoencoder-Level Set Method (AE-LSM) for ND in this paper. The AE structure is employed to reduce the feature space with high dimensionality to a 3-dimensional (3D) space. The LSM algorithm is trained based on the compressed 3D data to identify the boundary of normal data. The AE-LSM has advantages of boundary control and good generalization performance. Experiments on 5 benchmark UCI datasets and an Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset demonstrate that the proposed AE-LSM present a 3%~14% significant improvement based on the average AUC (p<0.05) over the AE and LSBD algorithms across the six datasets.

No abstract

The blood-oxygen label dependent (BOLD) signal obtained from functional magnetic resonance images (fMRI) varies significantly among populations. Yet, there is some agreement among researchers over the pace of the blood flow within several brain regions relative to the subject's age and cognitive ability. Our analysis further suggested that regional coherence among the BOLD fMRI voxels belonging to the individual region of the brain has some correlation with underlying pathology as well as cognitive performance, which can suggest potential biomarkers to the early onset of the disease. To capitalise on this we propose a method, called Regional Optimum Frequency Analysis (ROFA), which is based on finding the optimum synchrony frequency observed at each brain region for each of the resting-state BOLD frequency bands (Slow 5 (0.01-0.027Hz), Slow 4 (0.027-0.073Hz) and slow 3 (0.073 to 0.198Hz)), and the whole frequency band (0.01-0.167Hz) respectively. The ROFA is carried out on fMRI data of total 310 scans, i.e., 26, 175 and 109 scans from 21 young-healthy (YH), 69 elderly-healthy (EH) and 33 Alzheimer's disease (AD) patients respectively, where these scans include repeated scans from some subjects acquired at 3 to 6 months intervals. A 10-fold cross-validation procedure evaluated the performance of ROFA for classification between the YH vs EH, YH vs AD and EH vs AD subjects. Based on the confusion-matrix parameters; accuracy, precision, sensitivity and Matthew's correlation coefficient (MCC), the proposed ROFA classification outperformed the state-of-the-art Group-independent component analysis (Group-ICA), Functional-connectivity, Graph metrics, Eigen-vector centrality, Amplitude of low-frequency fluctuation (ALFF) and fractional amplitude of low-frequency fluctuations (fALFF) based methods with more than 94.99% precision and 95.67% sensitivity for different subject groups. The results demonstrate the effectiveness of the proposed ROFA parameters (frequencies) as adequate biomarkers of Alzheimer's disease.