Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging AI and Clinical Reasoning: Abductive Explanations for Alignment on Critical Symptoms

Belona Sonna, Alban Grastien

TL;DR

This work leverages formal abductive explanations, which offer consistent, guaranteed reasoning over minimal sufficient feature sets, to establish a robust framework for trustworthy AI in medical diagnosis.

Abstract

Artificial intelligence (AI) has demonstrated strong potential in clinical diagnostics, often achieving accuracy comparable to or exceeding that of human experts. A key challenge, however, is that AI reasoning frequently diverges from structured clinical frameworks, limiting trust, interpretability, and adoption. Critical symptoms, pivotal for rapid and accurate decision-making, may be overlooked by AI models even when predictions are correct. Existing post hoc explanation methods provide limited transparency and lack formal guarantees. To address this, we leverage formal abductive explanations, which offer consistent, guaranteed reasoning over minimal sufficient feature sets. This enables a clear understanding of AI decision-making and allows alignment with clinical reasoning. Our approach preserves predictive accuracy while providing clinically actionable insights, establishing a robust framework for trustworthy AI in medical diagnosis.

Bridging AI and Clinical Reasoning: Abductive Explanations for Alignment on Critical Symptoms

TL;DR

This work leverages formal abductive explanations, which offer consistent, guaranteed reasoning over minimal sufficient feature sets, to establish a robust framework for trustworthy AI in medical diagnosis.

Abstract

Artificial intelligence (AI) has demonstrated strong potential in clinical diagnostics, often achieving accuracy comparable to or exceeding that of human experts. A key challenge, however, is that AI reasoning frequently diverges from structured clinical frameworks, limiting trust, interpretability, and adoption. Critical symptoms, pivotal for rapid and accurate decision-making, may be overlooked by AI models even when predictions are correct. Existing post hoc explanation methods provide limited transparency and lack formal guarantees. To address this, we leverage formal abductive explanations, which offer consistent, guaranteed reasoning over minimal sufficient feature sets. This enables a clear understanding of AI decision-making and allows alignment with clinical reasoning. Our approach preserves predictive accuracy while providing clinically actionable insights, establishing a robust framework for trustworthy AI in medical diagnosis.
Paper Structure (40 sections, 4 theorems, 3 equations, 3 figures, 3 tables, 2 algorithms)

This paper contains 40 sections, 4 theorems, 3 equations, 3 figures, 3 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background: Critical Symptoms and Reasoning Alignment in Medical Diagnosis
  3. Preliminaries
  4. Understanding the Diagnosis of Patients: Clinicians vs AI Model
  5. Notation
  6. Definition of Abductive Explanations
  7. Making AI Models Reason as Clinicians
  8. Relevant Features in Diagnosis
  9. Critical Symptoms in Diagnosis
  10. Embedding Critical Symptoms in Model Reasoning
  11. Alignment Between Clinical Reasoning and AI Systems on Critical Cases
  12. The Basis of Alignment
  13. From Aligned Explanations to Aligned AI Systems
  14. Identification of Critical Properties within the Training Dataset
  15. Identifying Critical Features
  16. ...and 25 more sections

Key Result

Theorem 1

A critical case diagnosis is aligned if and only if all its abductive explanations are both critical and aligned.

Figures (3)

  • Figure 1: Patient Diagnosis: AI System vs Clinicians
  • Figure 2: Quantifying Misalignment on the Breast Cancer Dataset.
  • Figure 3: Quantifying Misalignment on the Heart Disease Dataset.

Theorems & Definitions (11)

  • Definition 1: Relevant Feature(s)
  • Definition 2: Critical Feature(s)
  • Definition 3: Critical Case
  • Definition 4: Critical Explanation
  • Definition 5: Aligned Critical Explanation
  • Theorem 1: Aligned Diagnosis
  • Theorem 2: Alignment of an AI System
  • Definition 6: Critical Knowledge
  • Theorem 3: Existence of Critical Properties
  • Definition 7: Model Misclassification
  • ...and 1 more