MRI abdominal fat segmentation with a novel VAT/SAT delineation algorithm: Otsu regression calibration and comparison to k-means and fuzzy c-means.

Introduction: Obesity is a major cardiovascular risk factor, particularly in excessive visceral adipose tissue accumulation. This study assessed the efficiency of three image processing algorithms (Otsu, K-means, and Fuzzy C-means) in quantifying abdominal adipose tissue from single-slice MRI image analysis obtained using a standard acquisition protocol. Methods: We developed a novel, open-source algorithm to delineate visceral and somatic adipose tissue by comparing the distance from the centroid of the largest segmented regions to the center of the image. Segmentation methods were evaluated against a manual reference. Results: The study included 68 patients (30 males and 38 females) aged between 8 and 84. All algorithms showed satisfactory accuracy, with Otsu thresholding consistently performing slightly better. Segmentation efficiency was analyzed within subgroups defined by gender, age, weight status, and diagnosis. Accuracy remained acceptable across subgroups, although male sex and higher weight status in adults were associated with superior results. Linear regression models were implemented to address errors concerning visceral and somatic adipose tissue quantification. Correlations between adipose tissue surface and BMI emerged, with significant differences in adipose tissue distribution across genders and age groups. Discussion: Our findings highlight the potential of MRI-based adipose tissue segmentation in cardiovascular risk stratification using a novel, open-source algorithm.