Abstracts

RETURN TO ABSTRACT LISTING


1743. Do Texture Features on Non-Contrast and Contrast-Enhanced CT Help Identify Adenomas in Adrenal Nodules More than 10 Hounsfield Units?
Authors * Denotes Presenting Author
  1. Richa Patel *; University of Washington
  2. Zi Jun Wu; University of Washington
  3. Toshimasa Clark; University of Colorado, Anschutz Medical Campus
  4. Daniel Hippe; University of Washington
  5. Erik Soloff; University of Washington
  6. Maggie Zhang; University of Michigan
  7. Carolyn Wang; University of Washington
Objective:
Adrenal nodules are commonly seen on computed tomography (CT) and may be difficult to characterize based on mean attenuation (MA) alone. Our purpose was to determine if texture features can assist MA in characterizing adrenal nodules on non-contrast CT (NC-CT) and contrast-enhanced CT (CE-CT).

Materials and Methods:
IRB-approved, HIPAA-compliant retrospective study evaluated > 1 cm and > 10 Hounsfield units (HU) adrenal nodules on both NC-CT and CE-CT. A region of interest (ROI) was drawn in each adrenal lesion around its greatest axial dimensions on a single CT slice. This ROI was analyzed with texture analysis software (RADIOMICS 1.2.0.0). Logistic regression was used to evaluate the association of each texture feature with a non-adenoma diagnosis with adjustment for mean attenuation. NC-CT and CE-CT features were analyzed separately. To account for testing 43 features, we used a Bonferroni-adjusted significance level of a = 0.00116 (0.05/43). A subset of this data has been previously presented but this submission now includes new analysis involving the utility of texture features on CE-CT in distinguishing adenomas and how these features compare on NC-CT.

Results:
There are 76 patients with 84 adrenal nodules with mean attenuation > 10 HU and > 1 cm on NC-CT, also seen on CE-CT within 6 months of NC-CT. Of 84 nodules, 29 were non-adenomas (primarily metastases) and 55 were presumed adenomas based on imaging or 12 month stability. Logistic regression analysis of 43 texture features on nodules from NC-CT demonstrated busyness (OR for non-adenoma: 7.2 per 1-SD increase, p=0.0001) and strength (OR: 0.2, p=0.0004) were significantly associated with a non-adenoma diagnosis, independent of MA. The analysis applied to the same nodules on CE-CT was consistent with similar associations of busyness (OR: 5.6, p=0.0031) and strength (OR = 0.2, p=0.0015) for non-adenomas, though these features did not quite reach the strict Bonferroni-adjusted significant threshold (p<0.00116). NC-CT busyness and CE-CT busyness were strongly correlated (rho=0.76) as were NC-CT strength and CE-CT strength (rho=0.78). Busyness and strength are inversely related (rho=-0.98) indicating adrenal lesions with relatively high busyness have relatively low strength, and vice versa.

Conclusion:
Second-order texture features busyness and strength can independently distinguish adenomas from non-adenomatous adrenal lesions on NC-CT. Adenomatous lesions have higher strength, while non-adenomatous lesions have higher busyness. A similar relationship occurs in CE-CT. While these features are inversely related to one another, both features correlate between NC-CT and CE-CT suggesting they can be used both in the presence and absence of contrast.