PURPOSE: To quantify specific characteristics of different types of ascitic fluid on magnetic resonance (MR) images and to determine their utility for computer-assisted lesion classification. METHODS: The MR images of 48 patients with intra-abdominal fluid were retrospectively analyzed. Patients were grouped according to the underlying disease and pathological outcomes. The fluid texture was analyzed on Breath Hold Axial T2 FatSat FIESTA sequence, using MaZda software. Most discriminative texture features for the classification of different types of ascites were selected based on Fisher coefficients (F) and the probability of classification error and average correlation coefficients (POE+ACC). Computer-assisted classification based on k-nearest-neighbor (k-NN) and artificial neural network (ANN) was performed and then accuracy, sensitivity and specificity were calculated. RESULTS: Adequate discriminative power for differentiating benign ascites from malignant ascites was achieved for two textural features, namely the Run Length Nonuniformity computed from both vertical and horizontal directions with 91.84% accuracy (sensitivity 100%; specificity 42.86%), and ten features for differentiating bland from hemorrhagic fluid with 90.00% accuracy (sensitivity 92.31%; specificity 85.71%), both for the ANN classifier. CONCLUSION: Texture analysis revealed several differences in signal characteristics of benign and malignant ascites. Computer-assisted pattern recognition algorithms may aid in the differential diagnosis of ascites types, especially in the early stages when there are few peritoneal modifications or when the cause is difficult to find.
PURPOSE: To quantify specific characteristics of different types of ascitic fluid on magnetic resonance (MR) images and to determine their utility for computer-assisted lesion classification. METHODS: The MR images of 48 patients with intra-abdominal fluid were retrospectively analyzed. Patients were grouped according to the underlying disease and pathological outcomes. The fluid texture was analyzed on Breath Hold Axial T2 FatSat FIESTA sequence, using MaZda software. Most discriminative texture features for the classification of different types of ascites were selected based on Fisher coefficients (F) and the probability of classification error and average correlation coefficients (POE+ACC). Computer-assisted classification based on k-nearest-neighbor (k-NN) and artificial neural network (ANN) was performed and then accuracy, sensitivity and specificity were calculated. RESULTS: Adequate discriminative power for differentiating benign ascites from malignant ascites was achieved for two textural features, namely the Run Length Nonuniformity computed from both vertical and horizontal directions with 91.84% accuracy (sensitivity 100%; specificity 42.86%), and ten features for differentiating bland from hemorrhagic fluid with 90.00% accuracy (sensitivity 92.31%; specificity 85.71%), both for the ANN classifier. CONCLUSION: Texture analysis revealed several differences in signal characteristics of benign and malignant ascites. Computer-assisted pattern recognition algorithms may aid in the differential diagnosis of ascites types, especially in the early stages when there are few peritoneal modifications or when the cause is difficult to find.
Authors: Turkey Refaee; Zohaib Salahuddin; Yousif Widaatalla; Sergey Primakov; Henry C Woodruff; Roland Hustinx; Felix M Mottaghy; Abdalla Ibrahim; Philippe Lambin Journal: J Pers Med Date: 2022-03-31