John Sahaya Rani Alex

Introduction: Alzheimer’s disease (AD) is responsible for many dementia-related deaths worldwide, and Mild Cognitive Impairment (MCI) is often its earliest clinical manifestation. Early detection of MCI is essential to initiate timely interventions that can slow progression and improve the patient’s quality of life. The emergence of diffusion tensor imaging (DTI) has made it a powerful neuroimaging modality for MCI detection, offering greater sensitivity to microstructural changes in white matter than basic structural imaging. Methods: A Multichannel Attention Residual Learning–DTI (MARL–DTI) architecture was designed with attention mechanisms and residual learning utilizing multiple DTI-derived diffusion metrics, including fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AxD). To overcome the scarcity of labelled neuroimaging data, the model was supplemented with a few-shot learning framework. The dataset comprised 239 DTI scans from ADNI was used to evaluate the model. Local Interpretable Model-Agnostic Explanations (LIME) assists in analysing the model’s decision-making process behaviour. Results: The proposed MARL-DTI achieved a classification accuracy of 92.76%, outperforming baseline DenseNet variants, StatFusion-FCNN, and Matching Networks. Validation on an unseen dataset of 36 ADNI subjects yielded consistent accuracy (91.67%), confirming the model’s reliability. LIME analysis was used to provide instance-level explanations, enabling qualitative assessment of the model’s decision-making behaviour by highlighting feature contributions for individual predictions. Discussion: These results demonstrate that the proposed MARL-DTI framework surpasses traditional deep learning architectures in the early identification of MCI using DTI data. The model consistently achieved strong performance while maintaining transparency, using LIME as a validation tool to examine its decision-making behaviour. These findings highlight the potential of combining multi-channel volumetric DTI data with explainable deep learning approaches to support reliable and clinically relevant early detection of Alzheimer’s disease.

Correction on: Usha GP and Alex JSR ( 2026) AQFormer: severity-aware transformer with aphasia-specific CAM for spoken keyword classification in aphasic speech. Front. Artif. Intell. 9:1786757. doi: 10.3389/frai.2026.1786757Funding statement incorrect In the published article, there was a mistake in the Funding statement. The author(s) declared that financial support was not received for this work and/or its publication. The correct statement reads: [The author(s) declared that financial support was received for this work. The APC was paid by V ellore Institute of Technology, Chennai, India.] The original version of this article has been updated.for a reason not seen here, please contact the journal's editorial office.

Language-driven speech output in individuals with aphasia shows considerable variability, including phonological errors and pauses during word searches. This makes it difficult to use traditional keyword classification systems and further reduces trust in deep neural models, complicating their application in clinical settings. This paper introduces AQFormer, a severity-aware transformer architecture designed to classify spoken keywords in aphasic speech, and A-CAM, a dual-stream attribute framework aimed at assisting individuals with aphasic impairments. AQFormer generates acoustic representations that are severity-adaptive by integrating patient-level Aphasia Quotient (AQ) scores through Feature-wise Linear Modulation (FiLM) and A-CAM. A-CAM consists of two main components: (i) a branch that influences WavLM convolutional features, a prediction-focused one, and (ii) a multimodal aphasia filter that captures pauses, phoneme variations, and interruptions at word boundaries, an impairment-focused branch. We introduce an adaptive perturbation and dual-filtering gradient scheme that enforces non-negative, mask-consistent attributions over time-frequency regions. Experiments utilizing a subset of AphasiaBank keywords (93 speakers, 960 recordings; training set expanded to 5,138) with rigorous speaker-disjoint evaluation indicate that AQFormer achieves approximately 96.61% accuracy (F1 = 96.8%) on previously unseen speakers. A-CAM consistently outperforms several Grad-CAM variants when deletion/insertion AUPC and ADCC metrics are employed. This results in stable, sparse explanations that reflect how aphasia is usually caused: Discriminates correct from incorrect productions with Cohen’s d = 2.05 (a massive effect size) and spatial localization of error regions with Intersection over Union (IoU) of 0.461 against phoneme boundaries. Montreal Forced Aligner meets the quantitative validation criteria for the aphasia filter. The impairment-focused A-CAM maps achieve an IoU of 0.712 against detected error regions, with a severity correlation that doubles from rho = −0.374 (base) to rho = −0.754 (filter-gated). By tightly coupling severity-aware modelling with aphasia-informed attributes, the proposed framework advances explainable learning systems for aphasia-affected speech without needing clinician-labelled training targets.

Cognitive and functional decline in Alzheimer’s disease (AD) arises from disruptions in specific brain networks. Identifying the most affected regions is essential for understanding disease progression and developing targeted interventions. Fluorodeoxyglucose positron emission tomography (FDG-PET) offers a sensitive method for detecting early metabolic dysfunction, often before structural changes become apparent. We examined regional brain glucose metabolism in relation to cognitive performance and functional independence across cognitively normal individuals, those with mild cognitive impairment (MCI), and AD patients. Cognitive function was measured using the Mini-Mental State Examination (MMSE), and daily functioning was assessed via the Functional Activities Questionnaire (FAQ). Imaging and clinical data were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI). Structural causal modeling was used to identify brain regions with a strong causal influence on MMSE and FAQ scores. These causally validated regions were then used as input to the proposed FDG-PET-based Cognition Prediction Network (FDG CogNet), a deep learning model, which includes a feature-wise attention mechanism to dynamically weight each region’s contribution to prediction. Temporal, parietal, and hippocampal regions were most influential for cognitive performance, particularly in early stages of disease. Functional abilities were more strongly associated with executive and integrative regions, including the angular gyrus, temporal poles, posterior cingulate, and frontal cortices. Cerebellar regions showed compensatory activity in MCI but diminished in AD, suggesting reduced neural resilience. FDG CogNet achieved high predictive accuracy, with R² = 0.90 for MMSE and R² = 0.94 for FAQ, demonstrating that limiting inputs to causally relevant regions improved both performance and interpretability. These findings clarify stage-specific neural mechanisms in AD and show that combining causal inference with an attention-based deep learning model provides a powerful framework for accurate and interpretable prediction. This approach highlights the clinical utility of FDG PET for early diagnosis and suggests that timely, region-specific interventions offer the best opportunity to preserve cognitive and functional abilities in AD.

Deep Neural Networks (DNNs) are one of the prominent and state-of-the-art methods for implementing artificial intelligence algorithms. A noticeable amount of effort has been put into hardware acceleration of DNN, with significantly less attention given to the SoftMax layer, which uses expensive hardware for exponentiation and division. The SoftMax function must be implemented as accurately as possible since it is crucial in multiclass classification training and inference tasks. This paper describes an efficient implementation of the SoftMax function for error-sensitive application of DNNs. We propose a Lookup Table (LUT) based on linear polynomial approximation for exponential and log units. The proposed architecture is modeled in Verilog and synthesized to gate-level netlist in GPDK180[ ]nm using Cadence[ ]. Results after the placement and routing stage show a delay of 7.15[ ]ns while consuming 5.3 mW of total power when operated at 100[ ]MHz. Multiclass classification is coded in Python with the proposed SoftMax layer to evaluate the accuracy in real-time applications. The results show that the neural network trained with the proposed SoftMax reaches almost the same accuracy as traditional SoftMax of 86.76%. This proves that low-form factor platforms can implement the same accuracy achieved with the proposed SoftMax function with less power.

Background: Biomarkers play an important role in the diagnosis of early-stage Alzheimer disease (AD), with the hippocampal emerging as the most reliable indicator of AD pathology. Elucidation of the patient’s left and right hippocampal volumes warrants further consideration. Therefore, caution should be exercised regarding the constraints inherent in the measurement method. This meta-analysis aimed to determine the left and right hippocampal volume changes and the volume measurement method. Methods: The PubMed, Web of Science, and Scopus databases were searched and published until September 2023. A total of 29 studies were included in this meta-analysis. Results: The meta-analysis results indicate a significant reduction in total hippocampal volume in AD patients. The results for the hippocampal volume on the left and right sides of 3 classes, such as AD, mild cognitive impairment (MCI), and normal control (NC), showed a greater reduction on the left side. The results showed in AD: 95% confidence interval (CI) = −142 to 1.61, P = .04, I ² = 44%; MCI: 95% CI= −186 to −72, P = <.00001, I ² = 0%, and NC: 95% CI = −141 to 14, P = .11, I ² = 80%, respectively, with random effect. The left hippocampal in AD is smaller, while the right hippocampal is larger. The AD class significantly reduced hippocampal volume on both the left and right sides. Conclusion: Notably, automatic segmentation methods impact hippocampal volume measurements, with results potentially relying heavily on their correlation with the manual labeling process.