Diffusion-driven SpatioTemporal Graph KANsformer for Medical Examination Recommendation

TL;DR

This work tackles medical examination recommendation by formulating it as a spatiotemporal, heterogeneous problem and addressing data noise with a diffusion-based denoising stage. The DST-GKAN architecture then applies a relation-aware spatial encoder (RGAT), a Temporal encoder (KANsformer) that leverages learnable spline activations, and a cross-attention fusion to produce robust patient representations for recommending examinations. The authors introduce MeExam, a temporally organized, heterogeneous dataset, to benchmark this task. Experiments show state-of-the-art performance over a wide range of baselines, with ablations confirming the contributions of diffusion denoising, spatiotemporal fusion, and the learned temporal dynamics. Overall, the approach advances automated diagnostic decision support by integrating noisy heterogeneous data into a coherent spatiotemporal model with practical clinical relevance.

Abstract

Recommendation systems in AI-based medical diagnostics and treatment constitute a critical component of AI in healthcare. Although some studies have explored this area and made notable progress, healthcare recommendation systems remain in their nascent stage. And these researches mainly target the treatment process such as drug or disease recommendations. In addition to the treatment process, the diagnostic process, particularly determining which medical examinations are necessary to evaluate the condition, also urgently requires intelligent decision support. To bridge this gap, we first formalize the task of medical examination recommendations. Compared to traditional recommendations, the medical examination recommendation involves more complex interactions. This complexity arises from two folds: 1) The historical medical records for examination recommendations are heterogeneous and redundant, which makes the recommendation results susceptible to noise. 2) The correlation between the medical history of patients is often irregular, making it challenging to model spatiotemporal dependencies. Motivated by the above observation, we propose a novel Diffusion-driven SpatioTemporal Graph KANsformer for Medical Examination Recommendation (DST-GKAN) with a two-stage learning paradigm to solve the above challenges. In the first stage, we exploit a task-adaptive diffusion model to distill recommendation-oriented information by reducing the noises in heterogeneous medical data. In the second stage, a spatiotemporal graph KANsformer is proposed to simultaneously model the complex spatial and temporal relationships. Moreover, to facilitate the medical examination recommendation research, we introduce a comprehensive dataset. The experimental results demonstrate the state-of-the-art performance of the proposed method compared to various competitive baselines.

Figures (12)

• Figure 1: Toy examples of medical examination recommendation. (a) comparison among medical examination recommendation and existing sequential recommendation methods. (b) Medical examination recommendation with heterogeneity property. Blue indicates diseases, gray represents symptoms red highlights the recommended examinations.
• Figure 2: The overview of proposed Diffusion-driven SpatioTemporal Graph KANsformers (DST-GKAN) model. It comprises two stages. (1) Task-adaptive Denoising: the task-adaptive diffusion model is employed for denoising. (2) Heterogeneous Spatiotemporal Representation Learning: Relation-aware Graph Attention Network (RGAT), KANsformer and Cross-Attention fusion are performed for feature representation.
• Figure 3: The illustration of forward (noising) and reverse (denoising) processes in our diffusion models.
• Figure 4: The framework of the proposed KANsformer. The KANs is integrated into the Transformer framework.
• Figure 5: The framework of MeExam: i) The few-shot NER framework NEEDLE is used to extract the medical heterogeneous entities required for patient sequences from anonymous medical notes text data. ii) Side information, which consists of the patient's age and gender, is collected through template matching or retrieval methods. iii) Employ large language models to cleanse, disambiguate, and standardize the identified heterogeneous entities from NEEDLE.