Decision processes in 3D structural MRI schizophrenia classification evaluated with saliency maps

2026-06-13 · Source: Machine learning : nature.com subject feeds · Field: Science & Research — Health & Medical Research, Mathematics & Computational Sciences, Research Methodology & Innovation · Depth: Expert, extended

Summary

This study evaluates seven deep learning (DL) architectures for classifying schizophrenia from 3D structural MRI data, focusing on the transparency of their decision processes using gradient-weighted class activation mapping (Grad-CAM). Researchers trained models on 192 adult brains (101 schizophrenia patients, 91 healthy controls) from the MCIC collection. While all models achieved over 70% classification accuracy and AUC scores above 0.75, only BrainID and ResNet18 demonstrated plausible decision-making based on structural brain information, as assessed by saliency maps. The study also developed a method to translate Grad-CAM saliency maps into universally interpretable anatomical markers, identifying candidate regions, primarily frontal and right-sided, consistent with known schizophrenia markers.

Key takeaway

For AI Scientists developing clinical decision support systems, you must integrate explainable AI (XAI) methods like Grad-CAM to validate model decisions, not just performance. Your models might achieve high accuracy but rely on implausible features, indicating overfitting or spurious correlations. Systematically evaluating saliency maps and deriving consistent anatomical markers will build trust and accelerate the clinical translation of deep learning applications, ensuring your diagnostic tools are both accurate and interpretable.

Key insights

Explainable AI methods are crucial for validating deep learning models in medical diagnostics, revealing decision plausibility beyond performance metrics.

Principles

• High classification accuracy does not guarantee plausible decision-making.
• Saliency maps can reveal overfitting or inconsistent feature learning.
• Aggregating local explanations yields global biomarker candidates.

Method

A three-stage approach: (1) train DL classifiers, (2) evaluate individual saliency map plausibility using Grad-CAM, mass accuracy, and center of mass, (3) derive global anatomical biomarkers by statistically comparing and intersecting consistent saliency regions across plausible architectures.

In practice

• Apply Grad-CAM to assess DL model decision plausibility in medical imaging.
• Quantify saliency map quality using mass accuracy and CoM deviation.
• Intersect saliency maps from multiple models for robust biomarker discovery.

Topics

• Deep Learning
• Schizophrenia Classification
• Explainable AI
• Grad-CAM
• Structural MRI
• Neuroimaging Biomarkers

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Med-CAM: Minimal Evidence for Explaining Medical Decision Making — Med-CAM generates minimal, sharp, and faithful evidence-based explanations for medical AI decisions.
• Learning Under Extreme Data Scarcity: Subject-Level Evaluation of Lightweight CNNs for fMRI-Based Prodromal Parkinsons Detection — Subject-level data splitting and model capacity are crucial for reliable deep learning evaluation in extreme data scarcity.
• SAIL: Structure-Aware Interpretable Learning for Anatomy-Aligned Post-hoc Explanations in OCT — Integrating anatomical priors and semantic features improves deep learning explainability in medical imaging.
• AI algorithm enables tracking of vital white matter pathways — An AI tool precisely segments brainstem white matter bundles in live MRI, revealing disease-specific structural changes and recovery.
• Dual-Modal Lung Cancer AI: Interpretable Radiology and Microscopy with Clinical Risk Integration — Integrating CT radiology and H&E microscopy with clinical data significantly enhances lung cancer diagnosis and interpretability.
• Bridging the interpretability gap for medical artificial intelligence models using class-association manifold learning — CAML enhances medical AI explainability by decoupling decision patterns and visualizing global class-associated knowledge.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine learning : nature.com subject feeds.