Body location embedded 3D U-Net (BLE-U-Net) for ovarian cancer ascites segmentation on CT scans

Abstract

Ascites is often regarded as the hallmark of advanced ovarian cancer, which is the most lethal gynecologic malignancy. Ascites segmentation contributes to track the progress of ovarian cancer development by providing accurate ascites measurement, which can effectively guide subsequent treatment and potentially reduce the mortality. Segmentation of ascites is challenging due to the presence of iso-intense fluids such as bile, urine, etc., near the ascites region. In this work we propose a novel 3D U-Net segmentation method called body location embedded U-Net (BLE-U-Net) that integrates anatomical location information with the segmentation process. BLE-U-Net incorporates body part regression to predict the approximate anatomical location of each CT slice along the z- axis. The regression scores are discretized to indicate different body regions and embedded into a modified 3D U-Net to improve the ascites segmentation. Twenty contrast-enhanced body CT scans were used to evaluate the proposed method. Dice coefficients of 38 ±10 and 65 ±06 were achieved for a conventional 3D U-Net and BLE-U-Net, respectively (with t-test p <0.05). Volumes of segmented ascites were 0.51±0.74 and 0.57±0.85 liters for each method where the ground-truth volume was 0.58±0.84 liters. These results suggest that the embedded location information is the key factor to improve the ascites segmentation, which could potentially benefit ovarian cancer diagnosis and treatment.

Deep‐learning based fast and accurate 3D CT deformable image registration in lung cancer

“…This might lead to worse agreements of DVH indices and a lower 3D Gamma passing rate (92.11% of 3 mm/3%/10% and 89.92% of 2 mm/2%/10% for the wMAE model and 92.30% of 3 mm/3%/10% and 90.21% of 2 mm/2%/10% for the M+S model, respectively). Moreover, for more challenging disease sites which have complexity shapes, such as ovarian cancer 94 and for registration between different modalities, 95 the proposed DIR approach may see its limitation. Further investigation is needed to address these issues.…”

Improved ascites segmentation with bladder identification using anatomical location residual U-Net

“…Urine in bladder is often misidentified as ascties in our previous work. 1 The addition of anatomical location information contributes to locatin the bladder, and two-class segmentation can explicitly identify bladder. Removing bladder regions from ascites segmentation can thus improve ascites segmentation accuracy, which is validated in our experimental results (Table 1).…”

“…The detailed information is referred to our previous work. 1 The class label of each slice is encoded into a feature vector F ∈ R W ×H×D×C with the same cardinality as the corresponding CT volume through an embedding layer, where W , H, D, and C are the slice width, height, depth, and channel, respectively.…”

“…We instead developed a 3D residual U-Net to keep such coherence by directly training on 3D CT volumes in our early work. 1 Moreover, anatomical location information was also embedded in 3D Residual U-Net based on the observation that ascites is accumulated in the abdomen but rarely resided in the chest. However, this method often includes the bladder filled w ith u rine i nto t he a scites s egmentation ( Figure 1 (C)) b ecause t hey h ave s imilar i ntensity v alues.…”