A Progressive Single-Modality to Multi-Modality Classification Framework for Alzheimer's Disease Sub-type Diagnosis

TL;DR

The paper addresses the cost and practicality of diagnosing Alzheimer's disease sub-types by proposing a progressive framework that starts with easily obtainable tabular data and incrementally incorporates MRI and PET data. It introduces a text disentanglement network to better extract information from textualized tabular data, a transformer-based multi-modality fusion module to integrate available modalities, and a progressive classifier that decides when to stop data acquisition based on a confidence threshold. Key innovations include a modality alignment loss $\mathcal{L}_{alig}$ to propagate late-stage information to early stages and a clinical guideline alignment via a contrastive loss $\mathcal{L}_{con}^k$ to enforce sub-type-specific adherence, enabling accurate sub-type classification under cost constraints. Experiments on 8280 subjects across four diverse datasets show the proposed approach outperforms state-of-the-art methods and achieves favorable cost-efficiency, with ablations validating the contributions of text disentanglement, alignment, and the progressive strategy. The work offers a practical, scalable path toward guideline-consistent, cost-aware AD sub-type diagnosis suitable for clinical deployment.

Abstract

The current clinical diagnosis framework of Alzheimer's disease (AD) involves multiple modalities acquired from multiple diagnosis stages, each with distinct usage and cost. Previous AD diagnosis research has predominantly focused on how to directly fuse multiple modalities for an end-to-end one-stage diagnosis, which practically requires a high cost in data acquisition. Moreover, a significant part of these methods diagnose AD without considering clinical guideline and cannot offer accurate sub-type diagnosis. In this paper, by exploring inter-correlation among multiple modalities, we propose a novel progressive AD sub-type diagnosis framework, aiming to give diagnosis results based on easier-to-access modalities in earlier low-cost stages, instead of modalities from all stages. Specifically, first, we design 1) a text disentanglement network for better processing tabular data collected in the initial stage, and 2) a modality fusion module for fusing multi-modality features separately. Second, we align features from modalities acquired in earlier low-cost stage(s) with later high-cost stage(s) to give accurate diagnosis without actual modality acquisition in later-stage(s) for saving cost. Furthermore, we follow the clinical guideline to align features at each stage for achieving sub-type diagnosis. Third, we leverage a progressive classifier that can progressively include additional acquired modalities (if needed) for diagnosis, to achieve the balance between diagnosis cost and diagnosis performance. We evaluate our proposed framework on large diverse public and in-home datasets (8280 in total) and achieve superior performance over state-of-the-art methods. Our codes will be released after the acceptance.

Figures (1)

• Figure 1: We propose a progressive multi-modality and multi-stage contrastive framework for AD sub-type diagnosis. Starting from Stage 1, we acquire the features from the textualized tabular data, and align them with the corresponding sub-type criteria in the guideline. In cases where the diagnosis confidence is low, we progressively utilize MR and PET images in Stages 2 and 3, respectively. To acquire modality information of the later stage(s) without actual data acquisition, we further align the features from early stages with those from the later stages.