Exploring the Feasibility of AI-Assisted Spine MRI Protocol Optimization Using DICOM Image Metadata

TL;DR

The paper addresses the challenge of protocol variability in spine MRI by leveraging AI trained on real-world DICOM metadata to optimize acquisition parameters. It uses four spine datasets (C-ST1, C-ST2, LS-ST1, LS-ST2) and a suite of classifiers (including RF and GB) with nested cross-validation, evaluating image-quality targets derived from $EntropyPower$ and $SpectralFlatness$ via F1. SHAP analyses reveal consistent, MRI-theory-aligned trends, notably involving $pFOV$, $FOV$, and $RT$, and demonstrate that the approach can identify actionable parameter patterns while offering interpretability. The findings indicate feasibility for AI-assisted protocol optimization and QC support in clinical practice, with potential for continuous learning and standardization across MRI systems, albeit with limitations related to dataset size and evaluation methodology. The work suggests a pathway toward reducing repeats and improving efficiency in clinical MRI through data-driven, interpretable protocol guidance.

Abstract

Artificial intelligence (AI) is increasingly being utilized to optimize magnetic resonance imaging (MRI) protocols. Given that image details are critical for diagnostic accuracy, optimizing MRI acquisition protocols is essential for enhancing image quality. While medical physicists are responsible for this optimization, the variability in equipment usage and the wide range of MRI protocols in clinical settings pose significant challenges. This study aims to validate the application of AI in optimizing MRI protocols using dynamic data from clinical practice, specifically DICOM metadata. To achieve this, four MRI spine exam databases were created, with the target attribute being the binary classification of image quality (good or bad). Five AI models were trained to identify trends in acquisition parameters that influence image quality, grounded in MRI theory. These trends were analyzed using SHAP graphs. The models achieved F1 performance ranging from 77% to 93% for datasets containing 292 or more instances, with the observed trends aligning with MRI theory. The models effectively reflected the practical realities of clinical MRI settings, offering a valuable tool for medical physicists in quality control tasks. In conclusion, AI has demonstrated its potential to optimize MRI protocols, supporting medical physicists in improving image quality and enhancing the efficiency of quality control in clinical practice.

Figures (9)

• Figure 1: SHAP values results for the features in the C-ST1 dataset.
• Figure 2: Top five most important features for each model, weighted by their respective F1-Score results for the C-ST1 dataset.
• Figure 3: SHAP values results for the features in the LS-ST1 dataset.
• Figure 4: Top five most important features for each model, weighted by their respective F1-Score results for the LS-ST1 dataset.
• Figure 5: SHAP values results for the features in the C-ST2 dataset.