Volumetric quantitative histogram analysis using diffusion-weighted magnetic resonance imaging to differentiate HCC from other primary liver cancers

Abstract

Objective

To evaluate the ability of volumetric quantitative apparent diffusion coefficient (ADC) histogram parameters and LI-RADS categorization to distinguish hepatocellular carcinoma (HCC) from other primary liver cancers [intrahepatic cholangiocarcinoma (ICC) and combined HCC-ICC].

Methods

Sixty-three consecutive patients (44 M/19F; mean age 62 years) with primary liver cancers and pre-treatment MRI including diffusion-weighted imaging (DWI) were included in this IRB-approved single-center retrospective study. Tumor type was categorized pathologically. Qualitative tumor features and LI-RADS categorization were assessed by 2 independent observers. Lesion volume of interest measurements (VOIs) were placed on ADC maps to extract first-order radiomics (histogram) features. ADC histogram metrics and qualitative findings were compared. Binary logistic regression and AUROC were used to assess performance for distinction of HCC from ICC and combined tumors.

Results

Sixty-five lesions (HCC, n = 36; ICC, n = 17; and combined tumor, n = 12) were assessed. Only enhancement pattern (p < 0.015) and capsule were useful for tumor diagnosis (p < 0.014). ADC 5th/10th/95th percentiles were significant for discrimination between each tumor types (all p values < 0.05). Accuracy of LI-RADS for HCC diagnosis was 76.9% (p < 0.0001) and 69.2% (p = 0.001) for both observers. The combination of male gender, LI-RADS, and ADC 5th percentile yielded an AUROC/sensitivity/specificity/accuracy of 0.90/79.3%/88.9%/81.5% and 0.89/86.2%/77.8%/80.0% (all p values < 0.027) for the diagnosis of HCC compared to ICC and combined tumors for both observers, respectively.

Conclusion

The combination of quantitative ADC histogram parameters and LI-RADS categorization yielded the best prediction accuracy for distinction of HCC compared to ICC and combined HCC-ICC.

from A via a.sfakia on Inoreader http://bit.ly/2Rvegj6