AI-Assisted Brain Tumor MRI Reporting and Treatment-Planning Segmentation: A Retrospective Paired Workflow Evaluation

Area of Science:

• Neuroradiology
• Medical Imaging Analysis
• Artificial Intelligence in Healthcare

Background:

• Brain tumor magnetic resonance imaging (MRI) reporting and segmentation are critical for treatment planning but are often time-consuming and variable.
• Evaluating the impact of AI assistance on the efficiency, consistency, and reproducibility of these tasks is essential.

Purpose of the Study:

• To assess the association between AI assistance and changes in efficiency, consistency, and reproducibility for brain tumor MRI reporting and segmentation.
• To quantify the impact of AI on task completion times and accuracy metrics.

Main Methods:

• A retrospective study involving 30 brain tumor MRI cases (vestibular schwannomas, meningiomas, brain metastases).
• Two neuroradiologists performed diagnostic reporting with and without AI assistance.
• Two physicians completed tumor segmentation with and without AI-generated contours, following a 3-week washout period.

Main Results:

• AI assistance significantly reduced reporting times for one reader (40.39% reduction) and contouring times for both readers (100.00% and 87.11% reductions).
• Report-similarity metrics (ROUGE-L, BERTScore F1, Sentence-BERT) significantly increased with AI assistance (p < 0.001).
• Contour overlap (Dice coefficients) numerically improved with AI assistance (from 0.81 to 0.87 and 0.83 to 0.87).

Conclusions:

• AI assistance is associated with improved efficiency in brain tumor MRI reporting and segmentation workflows.
• AI tools led to higher report similarity and numerically improved contour overlap.
• Prospective validation is needed to confirm if these workflow efficiencies translate to broader clinical benefits.