Machine learning-based texture analysis for differentiation of large adrenal cortical tumours on CT

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Differentiating adrenal adenomas from carcinomas is crucial for appropriate patient management.
• Conventional radiologist evaluation of adrenal masses can be challenging, particularly for large lesions.

Purpose of the Study:

• To compare the diagnostic efficacy of CT texture analysis with traditional radiologist assessment for distinguishing between large adrenal adenomas and carcinomas.
• To evaluate the potential of CT texture analysis to improve the accuracy of adrenal tumor classification.

Main Methods:

• Quantitative CT texture analysis was performed on 54 histopathologically proven adrenal masses.
• Intensity- and geometry-based textural features were extracted from precontrast and venous-phase CT images.
• A random forest classifier utilizing textural features and attenuation values was compared against radiologist classifications based on morphological criteria.

Main Results:

• The CT texture analysis model achieved a mean accuracy of 82%, significantly outperforming radiologists (68.5%, p<0.0001).
• The study demonstrated improved interobserver agreement with the predictive model compared to radiologist assessments.
• The Dice similarity coefficient indicated high agreement between radiologists' image labels (0.875±0.04).

Conclusions:

• CT texture analysis shows promise in enhancing the evaluation of adrenal cortical tumors.
• This quantitative imaging biomarker may improve the differentiation between benign adrenal adenomas and malignant carcinomas.
• Texture analysis offers a valuable tool for improving diagnostic accuracy in adrenal mass characterization.