Can Ultrasound Texture Analysis Differentiate Liver Metastases According to the Histopathological Origin of the Primary Tumor?

Abstract

Objective: We aimed to investigate whether ultrasound-based texture analysis can differentiate liver metastases according to the histopathological origin of the primary tumor and to evaluate the quantitative texture characteristics of metastases originating from colorectal, pancreatic, and breast cancer. Materials and Methods: This prospective study included 75 patients with biopsy-proven liver metastases, comprising 25 colorectal adenocarcinoma, 25 pancreatic ductal adenocarcinoma, and 25 invasive ductal breast carcinoma metastases. Conventional B-mode ultrasound images were obtained prior to treatment. The largest metastatic lesion in each patient was manually segmented using a whole-lesion two-dimensional region of interest (ROI). Histogram-based texture analysis was performed using an in-house MATLAB-based software package (version R2021a; MathWorks, Natick, MA, USA). Extracted parameters included intensity-based metrics, dispersion measures, entropy, uniformity, and percentile values. Texture features were compared among the three groups using appropriate statistical tests. Results: Significant differences were observed among metastatic lesions according to their primary tumor origin. Significant differences were observed in the mean, median, minimum, maximum, most frequent gray-level values, root-mean-square level, root-sum-of-squares level, entropy, and all evaluated percentile parameters among groups (all p < 0.05). Pancreatic cancer metastases consistently demonstrated the highest intensity-related histogram values and percentiles, whereas breast cancer metastases exhibited the lowest values. Colorectal metastases were generally of intermediate intensity. Entropy values were significantly higher in colorectal and pancreatic metastases than in breast cancer metastases (p < 0.05), suggesting greater structural heterogeneity. No significant differences were observed for kurtosis, skewness, uniformity, or size distribution parameters (all p > 0.05). Conclusions: Ultrasound-based tissue analysis revealed distinct quantitative features among liver metastases originating from colorectal, pancreatic, and breast cancer. Density-related parameters, percentiles, and entropy show the potential to differentiate metastatic lesions based on their primary tumor origin, thus serving as a non-invasive biomarker.