AB035. Mediastinal lesion characterization and prognostication using diffusion MRI: from whole lesion to target zone with dynamic approach

Le résumé fourni par la source

Background: Invasive surgical procedures are currently the cornerstone of mediastinal oncologic staging but can lead to patient discomfort and morbidity. In this study we attempted to use a noninvasive magnetic resonance (MR) imaging approach to differentiate benign from malignant mediastinal lesions and at the same time attempt to provide an imaging-based preoperative evaluation of lesion resectability and to use a slider-based software for b-value appointment for prognostication of outcome. Methods: Fifty consecutive patients (31 male, 19 female) with a suspicious mediastinal lesion on clinical or on imaging-based [computed tomography (CT), positron emission tomography (PET) or endobronchial ultrasound (EBUS)] examinations underwent a magnetic resonance imaging (MRI), including diffusion-weighted imaging (DWI), one day prior to surgery. All MR images were evaluated by visual inspection and by calculating the apparent diffusion coefficient (ADC) values (mean) of the whole lesion and of the suspected zone on b2000 DWI. Moreover, a dynamic slider-based software was used to find the extrapolated b-value where the lesion disappears in the background noise. Histological examinations of operative specimens served as reference. Finally, we tried to predict resectability and to make a statement about prognosis. Results: In total 19 benign and 31 malignant lesions were included in this study. Differentiating malignant from benign lesions based on whole-lesion mean ADC was acceptable (sensitivity/specificity/accuracy 87%/47%/72%, optimal threshold 2.08×10−3 mm2/s). On b2000 DWI were expected to only show malignant lesions. Surprisingly, using this cut-off value in the dynamic slider-based assessment of extrapolated DWI, differentiation between benign and malignant lesions showed remarkable sensitivity, specificity and accuracy of 100%, 68% and 88%. Survival analysis showed the group of cases with sliding b above 2,000 s/mm2 to have a much shorter overall survival compared to those below 2,000 (hazard ratio 13, likelihood ratio test P<0.001). A very high correlation was found between MR signs of local invasion and surgical resectability (Cohen’s kappa 0.88). Conclusions: DW-MRI offers valuable insights for preoperative characterization and resectability assessment of mediastinal lesions. Targeted lesion analysis and dynamic software interpretation hold promise for improved diagnostic accuracy Future integration of artificial intelligence may further enhance these capabilities.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.