Abstract
In this pre-use clinical evaluation study, we demonstrated clinical feasibility and usefulness as well as the potential impact of a previously suggested imaging biomarker for radiological assessment of Hippocampal Sclerosis (HS).
Introduction
Hippocampal Sclerosis (HS) represents the most frequent cause of drug-resistant epilepsy. Accurate identification of HS is a critical step in surgical planning, yet diagnosing the condition using standard 3T MRI remains a significant clinical challenge. Currently, diagnosis relies heavily on the subjective visual assessment of radiologists, who must identify subtle volume loss and signal intensity changes.
While ultra-high-field (7T) MRI offers superior resolution and diagnostic clarity, these scanners are expensive and not widely available. This often leaves clinicians with equivocal results from standard 3T scans. This study evaluates whether an automated, deep-learning-based quantitative biomarker—specifically the surface-to-volume ratio—can serve as a decision-support tool to improve diagnostic accuracy on standard 3T MRI.
Methodology
The researchers conducted a "pre-use" clinical evaluation involving 40 epilepsy patients scanned at Bern University Hospital (Inselspital) between 2019 and 2022. To establish a rigorous "ground truth," two senior experts reviewed all available clinical data, including high-resolution 7T MRI. The cohort was intentionally "difficult," consisting of patients referred for 7T imaging because their initial 3T scans were inconclusive.
Six raters, including three neurologists, two neuroradiologists, and one radiologist-in-training, evaluated the 3T scans across two distinct phases:
- Round 1 (MRI only): Raters performed assessments based on standard MRI images, simulating current clinical practice.
- Wash-out Period: A minimum 30-day gap was enforced to minimize the risk of raters remembering specific cases.
- Round 2 (MRI + QReport): Raters re-evaluated the same cases, but were provided with a Quantitative Report (QReport).
The QReport was generated using the DL+DiReCT deep-learning tool. Unlike traditional volumetric analysis, this tool calculates the surface-to-volume ratio of the hippocampi. As HS causes the hippocampus to atrophy and lose its internal structural complexity, this ratio changes predictably. The report visualizes these metrics against a normative distribution, providing an interpretable "why" behind the data.
Results
The introduction of the QReport led to a substantial improvement in diagnostic performance across all metrics:
- Diagnostic Accuracy: The mean accuracy across all six raters increased from 77.5% in Round 1 to 86.3% in Round 2. This improvement was statistically significant with a very large effect size ($d = 1.43$).
- Inter-rater Agreement: The consistency between different doctors (Fleiss’ kappa) rose from a moderate 0.56 to a substantial 0.72, suggesting that quantitative data helps harmonize clinical interpretations.
- Confidence and Error Reduction: All raters reported higher confidence in their Round 2 assessments. Notably, the QReport helped correct a tendency toward over-diagnosis; Round 1 saw 36 false positives, which were significantly reduced in Round 2.
Interestingly, neurologists—who typically have less specialized training in neuroimaging than neuroradiologists—showed the most significant shifts in their assessments when using the QReport, often leading to more accurate outcomes.
Discussion and Conclusion
This study highlights the potential of quantitative biomarkers to bridge the gap between standard 3T imaging and specialized 7T MRI. By providing an interpretable, objective metric rather than a "black-box" diagnosis, the tool supports clinical decision-making without replacing the expert's judgment.
While the study is limited by a relatively small sample size of raters and cases, the results are a strong proof-of-concept. Future work will focus on integrating these reports into live clinical workflows and expanding the quantitative analysis to other brain regions, such as the amygdala, to further refine the diagnostic pipeline for epilepsy surgery.
Summary updated