# Adrenal lesion classification revisited: validation and adjustment of dual-energy CT derived virtual unenhanced attenuation thresholds

## TL;DR

This study evaluates and adjusts thresholds for classifying adrenal lesions using dual-energy CT to improve accuracy in distinguishing between adenomas and metastases.

## Contribution

The paper provides a cross-validated threshold for adrenal lesion classification using dual-energy CT data.

## Key findings

- Previously reported VUE thresholds showed varying accuracy for adrenal lesion differentiation.
- A cross-validated threshold of 24.7 HU achieved a mean accuracy of 0.79.
- A threshold of 17.3 HU matched the specificity of the 10 HU unenhanced cutoff but with lower sensitivity.

## Abstract

Dual-energy CT (DECT)-derived virtual unenhanced (VUE) images have been investigated for adrenal lesion differentiation, yet previously reported thresholds vary, hampering clinical application. We aimed to test previous VUE thresholds for adrenal lesion differentiation in a large retrospective cohort, to provide a cross-validated threshold based on our data, and to investigate the influence of underlying malignancies on differentiation accuracy.

290 patients with 348 adrenal lesions (169 metastases, 179 adenomas) were included. Dual-layer DECT-derived VUE thresholds from 3 previous studies were retrieved, applied to our cohort and corresponding sensitivity/specificity/accuracy was calculated. Optimal threshold based on our data were determined using ROC-analysis with five-fold cross validation. Moreover, a threshold with similar specificity to the 10 HU threshold in unenhanced images was calculated. Subgroup analysis of adrenal lesion differentiation depending on underlying malignancies was performed.

The previously suggested thresholds were 20, 22 and 29 HU, and corresponding sensitivity/specificity/accuracy was 0.61/0.92/0.76, 0.67/0.91/0.78, and 0.82/0.59/0.71, respectively. The threshold determined from our cohort was 24.7 HU, yielding a sensitivity/specificity/accuracy of 0.76/0.81/0.79. Differentiation in disease-specific subgroups showed similar sensitivity/specificity/accuracy (Melanoma:0.78/0.84/0.79; Lung cancer:0.78/0.8/0.78; RCC:0.78/1/0.79). The VUE threshold to achieve a 0.98 specificity similar to the unenhanced 10 HU cutoff was 17.3 HU, yielding a sensitivity of 0.49.

Previous VUE attenuation thresholds showed a varying accuracy for differentiation between adenomas and metastases. A cross-validated VUE threshold of 24.7 HU yielded a mean accuracy of 0.79, whereas a threshold of 17.3 HU was best for achieving comparable specificity as reported for the 10 HU threshold in unenhanced images.