METHODS: We propose to use Residual Blocks with a 3 × 3 kernel size for local feature extraction and Non-Local Blocks to extract the global features. The Non-Local Block has the ability to extract global features without using a huge number of parameters. The key idea behind the Non-Local Block is to apply matrix multiplications between features on the same feature maps.
RESULTS: We trained and validated the proposed method on the LIDC-IDRI dataset which contains 1018 computed tomography scans. We followed a rigorous procedure for experimental setup, namely tenfold cross-validation, and ignored the nodules that had been annotated by
METHODS: We propose to use 3D Axial-Attention, which requires a fraction of the computing power of a regular Non-Local network (i.e., self-attention). Unlike a regular Non-Local network, the 3D Axial-Attention network applies the attention operation to each axis separately. Additionally, we solve the invariant position problem of the Non-Local network by proposing to add 3D positional encoding to shared embeddings.
RESULTS: We validated the proposed method on 442 benign nodules and 406 malignant nodules, extracted from the public LIDC-IDRI dataset by following a rigorous experimental setup using only nodules annotated by at least three radiologists. Our results show that the 3D Axial-Attention model achieves state-of-the-art performance on all evaluation metrics, including AUC and Accuracy.
CONCLUSIONS: The proposed model provides full 3D attention, whereby every element (i.e., pixel) in the 3D volume space attends to every other element in the nodule effectively. Thus, the 3D Axial-Attention network can be used in all layers without the need for local filters. The experimental results show the importance of full 3D attention for classifying lung nodules.
METHODS: In this work, we introduce a fully automated liver tumour segmentation approach in contrast-enhanced CT datasets. The method is a multi-stage technique which starts with contrast enhancement of the tumours using anisotropic filtering, followed by adaptive thresholding to extract the initial mask of the tumours from an identified liver region of interest. Localised level set-based active contours are used to extend the mask to the tumour boundaries.
RESULTS: The proposed method is validated on the IRCAD database with pathologies that offer highly variable and complex liver tumours. The results are compared quantitatively to the ground truth, which is delineated by experts. We achieved an average dice similarity coefficient of 75% over all patients with liver tumours in the database with overall absolute relative volume difference of 11%. This is comparable to other recent works, which include semiautomated methods, although they were validated on different datasets.
CONCLUSIONS: The proposed approach aims to segment tumours inside the liver envelope automatically with a level of accuracy adequate for its use as a tool for surgical planning using abdominal CT images. The approach will be validated on larger datasets in the future.
METHOD: The method based on constructing atlases for the portal and the hepatic veins bifurcations, the atlas is used to localize the corresponding vein in each segmented vasculature using atlas matching. Point-based registration is used to deform the mesh of atlas to the vein branch. Three-dimensional distance map of the hepatic veins is constructed; the fast marching scheme is applied to extract the centerlines. The centerlines of the labeled major veins are extracted by defining the starting and the ending points of each labeled vein. Centerline is extracted by finding the shortest path between the two points. The extracted centerline is used to define the trajectories to plot the required planes between the anatomical segments.
RESULTS: The proposed approach is validated on the IRCAD database. Using visual inspection, the method succeeded to extract the major veins centerlines. Based on that, the anatomic segments are defined according to Couinaud segmental anatomy.
CONCLUSION: Automatic liver segmental anatomy identification assists the surgeons for liver analysis in a robust and reproducible way. The anatomic segments with other liver structures construct a 3D visualization tool that is used by the surgeons to study clearly the liver anatomy and the extension of the cancer inside the liver.
METHODS: Knee image is first oversegmented to produce homogeneous superpixels. Then, a ranking model is developed to rank the superpixels according to their affinities to standard priors, wherein background superpixels would have lower ranking values. Finally, seed labels are generated on the background superpixel using Fuzzy C-Means method.
RESULTS: SAGE has achieved better interobserver DSCs of 0.94 ± 0.029 and 0.93 ± 0.035 in healthy and OA knee segmentation, respectively. Good segmentation performance has been reported in femoral (Healthy: 0.94 ± 0.036 and OA: 0.93 ± 0.034), tibial (Healthy: 0.91 ± 0.079 and OA: 0.88 ± 0.095) and patellar (Healthy: 0.88 ± 0.10 and OA: 0.84 ± 0.094) cartilage segmentation. Besides, SAGE has demonstrated greater mean readers' time of 80 ± 19 s and 80 ± 27 s in healthy and OA knee segmentation, respectively.
CONCLUSIONS: SAGE enhances the efficiency of segmentation process and attains satisfactory segmentation performance compared to manual and random walks segmentation. Future works should validate SAGE on progressive image data cohort using OA biomarkers.