Heart2Mind: Human-Centered Contestable Psychiatric Disorder Diagnosis System using Wearable ECG Monitors

TL;DR

This work presents Heart2Mind, a human-centered system for contestable psychiatric disorder diagnosis using wearable ECG monitors. It introduces a Multi-Scale Temporal-Frequency Transformer (MSTFT) that fuses time-domain and frequency-domain HRV features to classify RRI sequences with high accuracy (0.917 LOOCV, 0.891 5-fold CV; AUC up to 0.961). The CDI incorporates Self-Adversarial Explanations (SAEs) and contestable LLMs to detect explanation discrepancies and enable clinicians to contest or validate AI decisions, improving transparency and safety. Practically, this approach enables continuous, objective cardiac biomarker-based assessment while preserving clinical oversight through contestable AI, with open-source availability. The findings highlight the promise and remaining challenges of integrating wearables and contestable AI in psychiatric care, including the need for broader validation and domain-specific LLM optimization.

Abstract

Psychiatric disorders affect millions globally, yet their diagnosis faces significant challenges in clinical practice due to subjective assessments and accessibility concerns, leading to potential delays in treatment. To help address this issue, we present Heart2Mind, a human-centered contestable psychiatric disorder diagnosis system using wearable electrocardiogram (ECG) monitors. Our approach leverages cardiac biomarkers, particularly heart rate variability (HRV) and R-R intervals (RRI) time series, as objective indicators of autonomic dysfunction in psychiatric conditions. The system comprises three key components: (1) a Cardiac Monitoring Interface (CMI) for real-time data acquisition from Polar H9/H10 devices; (2) a Multi-Scale Temporal-Frequency Transformer (MSTFT) that processes RRI time series through integrated time-frequency domain analysis; (3) a Contestable Diagnosis Interface (CDI) combining Self-Adversarial Explanations (SAEs) with contestable Large Language Models (LLMs). Our MSTFT achieves 91.7% accuracy on the HRV-ACC dataset using leave-one-out cross-validation, outperforming state-of-the-art methods. SAEs successfully detect inconsistencies in model predictions by comparing attention-based and gradient-based explanations, while LLMs enable clinicians to validate correct predictions and contest erroneous ones. This work demonstrates the feasibility of combining wearable technology with Explainable Artificial Intelligence (XAI) and contestable LLMs to create a transparent, contestable system for psychiatric diagnosis that maintains clinical oversight while leveraging advanced AI capabilities. Our implementation is publicly available at: https://github.com/Analytics-Everywhere-Lab/heart2mind.

Figures (14)

• Figure 1: Evolution from Human-centered XAI toward Contestable AI Systems.
• Figure 2: The overview of Heart2Mind framework: (a) Cardiac Monitoring Interface (CMI), (b) Contestable Diagnosis Interface (CDI), and (c) the design of contestable LLMs system.
• Figure 3: Sequence diagram illustrating the cardiac signal recording workflow in the Cardiac Monitoring Interface (CMI). The diagram shows the interaction flow between user actions, web interface components, and Polar device communication, including device-specific branching for H9 (HR/RRI only) and H10 (ECG/HR/RRI) configurations.
• Figure 4: The dashboard of the Cardiac Monitoring Interface (CMI) showing two main panels: (a) User controls and device status indicators for clinical monitoring, (b) Cardiac signals displaying ECG signal, HR, and RRI time series.
• Figure 5: Architecture overview of the Multi-Scale Temporal-Frequency Transformer (MSTFT). The model processes raw RRI time series through parallel temporal and frequency pathways, integrating features via cross-attention fusion before final classification. Key components include noise-augmented input preprocessing, multi-scale temporal convolutions with stochastic skips, wavelet-based frequency transforms, and attention-based feature fusion blocks.