A Cross Attention Approach to Diagnostic Explainability using Clinical Practice Guidelines for Depression

TL;DR

A domain-general architecture called ProcesS knowledge-infused cross ATtention (PSAT) that incorporates clinical practice guidelines (CPG) when computing attention is proposed, which surpasses the performance of twelve baseline models and can provide explanations where other baselines fall short.

Abstract

The lack of explainability using relevant clinical knowledge hinders the adoption of Artificial Intelligence-powered analysis of unstructured clinical dialogue. A wealth of relevant, untapped Mental Health (MH) data is available in online communities, providing the opportunity to address the explainability problem with substantial potential impact as a screening tool for both online and offline applications. We develop a method to enhance attention in popular transformer models and generate clinician-understandable explanations for classification by incorporating external clinical knowledge. Inspired by how clinicians rely on their expertise when interacting with patients, we leverage relevant clinical knowledge to model patient inputs, providing meaningful explanations for classification. This will save manual review time and engender trust. We develop such a system in the context of MH using clinical practice guidelines (CPG) for diagnosing depression, a mental health disorder of global concern. We propose an application-specific language model called ProcesS knowledge-infused cross ATtention (PSAT), which incorporates CPGs when computing attention. Through rigorous evaluation on three expert-curated datasets related to depression, we demonstrate application-relevant explainability of PSAT. PSAT also surpasses the performance of nine baseline models and can provide explanations where other baselines fall short. We transform a CPG resource focused on depression, such as the Patient Health Questionnaire (e.g. PHQ-9) and related questions, into a machine-readable ontology using SNOMED-CT. With this resource, PSAT enhances the ability of models like GPT-3.5 to generate application-relevant explanations.

Figures (4)

• Figure 1: Phrase Extraction and Other Resource Generation for Explainable Depression Detection. The red circles containing C1, C2, and C3 denote our contributions. C1 represents the adaptation of the CLEF eRisk dataset; C2 shows the development of PHQ-9-based depression ontology, C3.a displays part 'a' our third contribution of this work, which is the development of depression-specific phrase embedding matrix, and C3.b is the knowledge-infused cross-attention network for explainable depression detection.
• Figure 2: Overview of PSAT model. Nine cross-attention (CA) blocks represent the Nine PHQ-9 questions. $n$ represents the number of topical phrases in a user document to map to the PHQ-9 ontology. $d$ is the embedding. PSAT allows visualization of PHQ-9-level attentions (represented in different colors) as application-relevant explanations useful for MHPs.
• Figure 3: The AKC Scores attained by PSAT on PRIMATE (PRIM), CLEF e-Risk (CLEF), and CAMS demonstrate the impact of cross-attention blocks. PSAT's higher range of AKC scores indicates its ability to utilize cross-attention effectively to exclude less relevant PHQ-9 questions, thus influencing the quality of explanations and the results.
• Figure 4: Illustration of the process followed by PHQ-9 Class Visualizer in mapping attention words of PSAT to PHQ-9-DO. The SNOMED-CT IDs in round boxes demonstrate a trace to clinically grounded concepts. These mapped concepts are then utilized to produce explanations as presented in Table \ref{['tab:hlex']}. Additionally, a contrast is drawn by showcasing ClinicalT5's attention, which highlights the entire text and gives low weight to important clinical concepts. More examples of explanations are provided in supplementary material from CLEF eRisk (page 2), CAMS (pages: 3-7), and PRIMATE (pages: 8-11).