# Alzheimer's disease

Using ADReSSo and three large self-supervised speech backbones, the study applies controlled noise and reverberation interventions to participant-speech-only, non-speech, and full-recording audio, and combines layer-wise linear decoding, input- and representation-space interventions, and geometric alignment analysis to show that controlled acoustic interventions alter Alzheimer's disease predictions across all three backbones, that noise, despite showing no significant diagnostic-group difference in the original data, produces the strongest intervention effects, and that these effects are systematically structured relative to the classifier's decision direction, replicate on the held-out test set, and reverse when the representation-space intervention direction is reversed.

Using ADReSSo and three large self-supervised speech backbones, the study applies controlled noise and reverberation interventions to participant-speech-only, non-speech, and full-recording audio, and combines layer-wise linear decoding, input- and representation-space interventions, and geometric alignment analysis to show that controlled acoustic interventions alter Alzheimer's disease predictions across all three backbones, that noise, despite showing no significant diagnostic-group difference in the original data, produces the strongest intervention effects, and that these effects are systematically structured relative to the classifier's decision direction, replicate on the held-out test set, and reverse when the representation-space intervention direction is reversed.

Addressing the difficulty of characterising depression in Alzheimer's disease (AD), this study built an explainable multimodal framework integrating linguistic and acoustic features from elicited picture-description speech, used a source-informed cross-dataset transfer strategy to compensate for the absence of directly observed depression labels in the target AD cohort, identified two distinct speech sub-phenotypes via unsupervised clustering, found that one cluster showed mathematical proximity (cosine similarity in a shared latent space) to the depression centroid of a depression-labelled source corpus, then validated cluster separability with a proxy supervised classification task in which the retained multimodal artificial neural network achieved 92.

Addressing the difficulty of characterising depression in Alzheimer's disease (AD), this study built an explainable multimodal framework integrating linguistic and acoustic features from elicited picture-description speech, used a source-informed cross-dataset transfer strategy to compensate for the absence of directly observed depression labels in the target AD cohort, identified two distinct speech sub-phenotypes via unsupervised clustering, found that one cluster showed mathematical proximity (cosine similarity in a shared latent space) to the depression centroid of a depression-labelled source corpus, then validated cluster separability with a proxy supervised classification task in which the retained multimodal artificial neural network achieved 92.

The work proposes M²PFN, an end-to-end framework that back-propagates task gradients into 3D-MRI and tabular encoders through differentiable inference, aligns the two modalities into a shared subspace matched to the in-context learning prior via disentanglement and a contrastive objective, and folds in a frozen tabular-only prediction through a learnable gated shortcut; on ADNI (n=2240, three-class CN/MCI/AD) it attains 65.55% macro-F1 and 82.21% macro-AUC, surpassing the compared unimodal and multimodal baselines, regresses baseline MMSE by swapping only the head (1250-subject sub-cohort, test MAE 1.743), and achieves the best AUC and lowest MMSE MAE across all baselines on two external cohorts (OASIS-3 and SCAN) with no retraining, transferring even when the cognitive instrument changes.

The work proposes M²PFN, an end-to-end framework that back-propagates task gradients into 3D-MRI and tabular encoders through differentiable inference, aligns the two modalities into a shared subspace matched to the in-context learning prior via disentanglement and a contrastive objective, and folds in a frozen tabular-only prediction through a learnable gated shortcut; on ADNI (n=2240, three-class CN/MCI/AD) it attains 65.55% macro-F1 and 82.21% macro-AUC, surpassing the compared unimodal and multimodal baselines, regresses baseline MMSE by swapping only the head (1250-subject sub-cohort, test MAE 1.743), and achieves the best AUC and lowest MMSE MAE across all baselines on two external cohorts (OASIS-3 and SCAN) with no retraining, transferring even when the cognitive instrument changes.