Automatic segmentation of thrombosed aortic dissection in post‐operative CT‐angiography images

Abstract

Purpose The thrombus in the false lumen (FL) of aortic dissection (AD) patients is a meaningful indicator to determine aortic remodeling but difficult to measure in clinic. In this study, a novel segmentation strategy based on deep learning was proposed to automatically extract the thrombus in the FL in post‐operative computed tomography angiography (CTA) images of AD patients, which provided an efficient and convenient segmentation method with high accuracy. Methods A two‐step segmentation strategy was proposed. Each step contained a convolutional neural network (CNN) to segment the aorta and the thrombus, respectively. In the first step, a CNN was used to obtain the binary segmentation mask of the whole aorta. In the second step, another CNN was introduced to segment the thrombus. The results of the first step were used as additional input to the second step to highlight the aorta in the complex background. Moreover, skip connection attention refinement (SAR) modules were designed and added in the second step to improve the segmentation accuracy of the thrombus details by efficiently using the low‐level features. Results The proposed method provided accurate thrombus segmentation results (0.903 ± 0.062 in dice score, 0.828 ± 0.092 in Jaccard index, and 2.209 ± 2.945 in 95% Hausdorff distance), which showed improvement compared to the methods without prior information (0.846 ± 0.085 in dice score) and the method without SAR (0.899 ± 0.060 in dice score). Moreover, the proposed method achieved 0.967 ± 0.029 and 0.948 ± 0.041 in dice score of true lumen (TL) and patent FL (PFL) segmentation, respectively, indicating the excellence of the proposed method in the segmentation task of the overall aorta. Conclusions A novel CNN‐based segmentation framework was proposed to automatically obtain thrombus segmentation for thrombosed AD in post‐operative CTA images, which provided a useful tool for further application of thrombus‐related indicators in clinical and research application.

Automatic Segmentation of Teeth, Crown–Bridge Restorations, Dental Implants, Restorative Fillings, Dental Caries, Residual Roots, and Root Canal Fillings on Orthopantomographs: Convenience and Pitfalls

“…There are several more reliable metrics, such as pixel accuracy and intersection-over-union (the Jaccard index); however, we used the DSC as it is not only a measure of how many positives were found but it also penalizes for false positives. Additionally, theory states that the DSC and Jaccard index approximate each other relatively and absolutely [ 97 , 98 ]. Thirdly, although the dataset contained OPGs of participants that have different nationalities and ethnic backgrounds we collected the data from a single center, which probably had a negative effect on the generalizability of the model [ 76 ].…”

Deep learning-based multi-stage postoperative type-b aortic dissection segmentation using global-local fusion learning

“…† Results of other models using 'Fusion learning' showing significant differences (p < 0.05) compared with the ones of 'Ours + Fusion learning'. segmentation methods, and a postoperative descending aorta segmentation method for type-a AD (Feng et al 2022a). All these methods can be applied to the postoperative scenario to achieve the segmentation of multiple subregions including TH and were evaluated using the same dataset in section 3.1 and the configurations in section 3.2.…”

“…Vascular volume quantification To verify the value and advantages of our framework in clinical application, we quantified the volumes of TL, FL, and TH for two patient groups: (1) the patients in group A with fully metabolized TH near the stent-graft and unexpanded FL (good outcome), and (2) the ones in group B with enlarged TH+FL three months after TEVAR (poor outcome). The volumes were calculated from the automated segmentations (Auto) (generated by our framework and the published postoperative segmentation model (Feng et al 2022a)) and the manual annotations (Manual) according to the real image resolution. The results are shown in figure 10.…”

“…Although all the mentioned-above studies focus on the preoperative type-b AD segmentation, they are important references when dealing with similar problems in the postoperative scenario. Recently, Feng et al (2022a) proposed a multistage postoperative AD segmentation framework for TL, FL, and TH using attention mechanism. However, it was designed for type-a AD treated by total arch replacement surgery and only focused on the descending aorta in the local image region that was manually defined.…”

“…Bland-Altman analyses of vascular volume quantifications for two groups of patients with good outcomes (Group A) and poor outcomes (Group B). Auto: volume quantifications for the automated segmentation results generated by our framework and the model published byFeng et al (2022a). Manual: volume quantifications for manual annotations.…”