How Sensitive Are Radiomic AI Models to Acquisition Parameters?

2026-05-14 · Source: Artificial Intelligence · Field: Health & Wellbeing — Medical Devices & Health Technology, Clinical Care & Medical Practice, Artificial Intelligence & Machine Learning · Depth: Advanced, quick

Summary

A new framework quantifies the sensitivity of radiomic AI models to varying CT scan acquisition parameters, a critical factor hindering clinical deployment due to performance drops across heterogeneous multicenter protocols. This work applies a mixed-effects framework to analyze the influence of clinically relevant parameters on model performance, accounting for subject-level random effects. Researchers applied this framework to lung cancer diagnosis using CT scans from two independent multicenter datasets, including a public database and proprietary data, and evaluated several architectures. The study identified an optimal CT configuration—X-ray tube current >= 200 mA, spiral pitch <= 1.5, and slice thickness <= 1.25 mm—which improved sensitivity from 0.79±0.04 to 0.90±0.10 and specificity from 0.47±0.10 to 0.79±0.13, balancing diagnostic quality with reduced radiation dose.

Key takeaway

For AI Engineers developing radiomic systems for clinical deployment, understanding and mitigating performance variability due to acquisition parameters is crucial. Your models will achieve significantly better cross-dataset robustness and diagnostic accuracy by adhering to optimized CT scan settings like an X-ray tube current of >= 200 mA, spiral pitch <= 1.5, and slice thickness <= 1.25 mm. This directly impacts model reliability and patient safety, making these parameters key considerations in your data acquisition and model training pipelines.

Key insights

Quantifying CT acquisition parameter sensitivity improves radiomic AI model robustness and clinical deployment.

Principles

• Heterogeneous acquisition protocols degrade AI performance.
• Mixed-effects models quantify parameter influence.
• Optimal parameters balance quality and dose.

Method

A mixed-effects framework quantifies acquisition parameter influence on radiomic AI model performance, accounting for subject-level random effects, applied to lung cancer diagnosis in CT scans.

In practice

• Target X-ray tube current >= 200 mA.
• Maintain spiral pitch <= 1.5.
• Use slice thickness <= 1.25 mm.

Topics

• Radiomic AI Models
• CT Acquisition Parameters
• Multicentre Protocols
• Lung Cancer Diagnosis
• Mixed-Effects Framework

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

Related on AIssential

• [R] Seeing arxiv endorser (eess.IV or cs.CV) CT lung nodule AI validation preprint — CT slice thickness significantly impacts AI lung nodule detection sensitivity more than dose reduction.
• Acquisition state behaves as a structured, measurable variable governing lung-nodule AI: kernel-driven measurement instability and noise-driven detection fragility, invisible to DICOM metadata — CT acquisition state, uncaptured by DICOM, critically dictates medical AI performance and failure modes.
• Parameter-Efficient Adaptation of SAM 3 for Automated ITV Generation from 4DCT Images — Parameter-efficient fine-tuning of SAM 3 with temporal coherence effectively automates ITV generation from 4DCT images using minimal data.
• using .npy dataset with 3D models [R] — Vague technical questions yield unhelpful answers; specificity is crucial for meaningful support.
• When is 3D Worth It? A Resource-Performance Frontier for CNNs and Transformers in Lung CT — Input dimensionality significantly impacts model performance and stability in volumetric medical imaging.
• Vision-language models for chest radiography do not always need the image — Many medical vision-language models for chest radiography achieve high accuracy by exploiting text-based finding-name priors rather than actual image interpretation.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.