Deep Learning Segmentation of Ascites on Abdominal CT Scans for Automatic Volume Quantification

TL;DR

The proposed deep learning method performed well in segmenting and quantifying the volume of ascites in concordance with expert radiologist assessments and showed strong agreement with expert assessments.

Abstract

Purpose: To evaluate the performance of an automated deep learning method in detecting ascites and subsequently quantifying its volume in patients with liver cirrhosis and ovarian cancer. Materials and Methods: This retrospective study included contrast-enhanced and non-contrast abdominal-pelvic CT scans of patients with cirrhotic ascites and patients with ovarian cancer from two institutions, National Institutes of Health (NIH) and University of Wisconsin (UofW). The model, trained on The Cancer Genome Atlas Ovarian Cancer dataset (mean age, 60 years +/- 11 [s.d.]; 143 female), was tested on two internal (NIH-LC and NIH-OV) and one external dataset (UofW-LC). Its performance was measured by the Dice coefficient, standard deviations, and 95% confidence intervals, focusing on ascites volume in the peritoneal cavity. Results: On NIH-LC (25 patients; mean age, 59 years +/- 14 [s.d.]; 14 male) and NIH-OV (166 patients; mean age, 65 years +/- 9 [s.d.]; all female), the model achieved Dice scores of 0.855 +/- 0.061 (CI: 0.831-0.878) and 0.826 +/- 0.153 (CI: 0.764-0.887), with median volume estimation errors of 19.6% (IQR: 13.2-29.0) and 5.3% (IQR: 2.4-9.7) respectively. On UofW-LC (124 patients; mean age, 46 years +/- 12 [s.d.]; 73 female), the model had a Dice score of 0.830 +/- 0.107 (CI: 0.798-0.863) and median volume estimation error of 9.7% (IQR: 4.5-15.1). The model showed strong agreement with expert assessments, with r^2 values of 0.79, 0.98, and 0.97 across the test sets. Conclusion: The proposed deep learning method performed well in segmenting and quantifying the volume of ascites in concordance with expert radiologist assessments.

Figures (10)

• Figure 1: Flow Diagram detailing excluded scans for each dataset used in this retrospective study.
• Figure 2: Confusion matrix for deep learning based ascites detection compared with radiologist assessments on test datasets. Volume >50 mL was considered to be positive for ascites. (Left) NIH-OV (n=166), (Right) UofW-LC (n=124).
• Figure 3: Violin plots of Dice score distribution as well as $r^2$ and Bland-Altman plots of automatic ascites volume measurement using the deep learning (DL) model versus radiologists for each test set; (a) NIH-LC, (b) NIH-OV, (c) UofW-LC. Note: For $r^2$ plots, the axes are in log scale to account for the volume of ascites that spans several orders of magnitude.
• Figure 4: Manual vs deep learning-based automatic measurement of ascites on contrast-enhanced CT images of an 85-year-old male patient with liver cirrhosis from NIH-LC. (a) Top Row: Manual annotation (red), Bottom Row: automatic annotation (blue), Left-to-Right: axial, sagittal, and coronal viewing planes. An exemplary case, showcasing the segmentation performance of the DL model on an 85-year-old male patient with liver cirrhosis. The scan was acquired with contrast, has a slice thickness of 5mm. As delineated by the arrows, the DL model selectively excluded specific regions that mimic ascites: the gallbladder (red arrow), pleural effusion (yellow arrow), and fluids within the stomach (green arrow). Additionally, the bowels were disregarded, as indicated by the blue arrow. (b) 3D rendering of ascites volume: manual measurement (red) vs automatic measurement (blue). The model achieved a segmentation Dice score of 0.920, with 13.3% volume estimation error (4.07L actual compared with 3.59L predicted).
• Figure 5: Manual vs deep learning-based automatic measurement of ascites on contrast-enhanced CT images of a 54-year-old female patient from NIH-LC. The images demonstrate an uncommon case of loculated ascites (yellow arrow). The scan was acquired with a 5mm slice thickness. The model achieved a segmentation Dice score of 0.717. With an estimated volume of 0.60L compared to the true volume of 0.76L, the model resulted in a 21.7% volume estimation error. Top Row: Manual annotation (red), Bottom Row: automatic annotation (blue), Left-to-Right: axial, sagittal, and coronal viewing planes.