Adipose-Joint Crosstalk in Obesity-Induced Osteoarthritis: Mechanisms, Biomarkers, and Translational Therapeutic Opportunities.

Alzheimer’s disease (AD) is the most common form of dementia, which is a loss of memory and cognitive function and a change in mood or personality []. It targets primarily the hippocampus, amygdala and other structures of the limbic system that play a key role in memory, emotions and other functions. cognitive function []. While many patients are initially found to have mild cognitive impairment (MCI), an intermediate state between normal cognition and AD, more severe symptoms develop later []. MCI is a major factor in the diagnosis of AD []. Diagnoses of AD can be made through a combination of clinical history, physical examinations, diagnostic tests, neurological examination, and the Mini-Mental State Examination (MMSE) [,]. But these heuristic tests are time consuming and may have inconsistent validity. AD specifically affects gray matter [] and neuroimaging techniques serve to complement analysis of brain function in AD []. Artificial neural networks (ANNs), models of information processing based on artificial neurons, have demonstrated great potential for early diagnosis of AD. Using data from neuroimaging, ANNs can detect different patterns and signals related to the disorder [,]. Following sufficient training on a large collection of neuroimages, these networks can classify previously unseen data, which will greatly reduce the time taken for patients to manage and assist clinicians with the accuracy of the AD diagnosis and monitoring. The field of artificial intelligence (AI) has made great strides in areas such as natural language processing, computer vision, and autonomous systems. AI has also shown potential in medicine for diagnostics, imaging analysis and disease prediction, with its ability to make fast, accurate and repeatable evaluations from complex data [,,]. Neural networks, fuzzy logic and hybrid models (e.g., fuzzy neural networks) [,] provide valuable contributions to the AD diagnostic toolbox. Many of the above methods solve the problems of traditional methods, such as objective, quantitative analysis of complex multimodal data, especially for imaging. Moreover, they are effective in managing uncertainty of diagnosis, facilitate early diagnosis and assist in personalized treatment strategies [,,,,,]. Even with data quality, interpretability, and clinical implementation issues persisting, AI shows significant promise in improving the precision, effectiveness, and promptness of Alzheimer’s diagnosis. This study was a feasibility study to assess a proposed classification framework based on MRI images of a publicly available Alzheimer’s disease dataset. Alzheimer’s stage images are categorized into four stages of Alzheimer’s disease: Non-Demented, Very Mild Demented, Mild Demented, and Moderate Demented []. These categories enable the evaluation of classification models in different stages of the disease and provide a useful reference for the evaluation of automated tools for Alzheimer’s staging. Although great progress has been made in learning for Alzheimer’s classification, there are still some practical problems. Deep convolutional neural networks have been used to achieve impressive results in medical image analysis, but they usually require huge amounts of annotated data, which may not be available in medical contexts []. Moreover, many deep learning models are black-box models, which are difficult to understand the reasons behind the prediction of individual models. This is especially crucial for clinical decision support in healthcare environments, where transparency, interpretability, and accountability are paramount []. Moreover, the dimensionality of the deep feature representations can lead to redundant representation of information, which could prompt the exploration of dimensionality reduction methods to build more compact and discriminative feature spaces. Previous works have reported encouraging results on classification performance based on deep learning models, but many fewer works have focused on incorporation of lightweight deep feature extraction, dimensionality reduction and adaptive fuzzy learning models for multi-stage classification of Alzheimer’s disease. There have been many studies employing deep learning on the problem of Alzheimer’s disease classification; but few studies have focused on combining lightweight deep feature extraction with adaptive classification mechanisms. In this study, a hybrid approach is explored based on deep feature extraction, dimensionality reduction and adaptive machine learning. The pre-trained SqueezeNet model is chosen due to its lightweight architecture and computational efficiency and is used to extract deep representations. Principal Component Analysis (PCA) is then used to decrease the number of features to the most useful variance. The compact feature representation obtained is then tested with diverse machine learning classifiers and the Enhancement Fuzzy Min–Max Neural Network (EFMM) is highlighted for its superiority. Unlik