Integration of spatial information in convolutional neural networks for automatic segmentation of intraoperative transrectal ultrasound images

Meta-Learning Initializations for Interactive Medical Image Registration

We present a meta-learning framework for interactive medical image registration. Our proposed framework comprises three components: a learning-based medical image registration algorithm, a form of user interaction that refines registration at inference, and a meta-learning protocol that learns a rapidly adaptable network initialization. This paper describes a specific algorithm that implements the registration, interaction and meta-learning protocol for our exemplar clinical application: registration of magnetic resonance (MR) imaging to interactively acquired, sparsely-sampled transrectal ultrasound (TRUS) images. Our approach obtains comparable registration error (4.26 mm) to the best-performing non-interactive learning-based 3D-to-3D method (3.97 mm) while requiring only a fraction of the data, and occurring in real-time during acquisition. Applying sparsely sampled data to non-interactive methods yields higher registration errors (6.26 mm), demonstrating the effectiveness of interactive MR-TRUS registration, which may be applied intraoperatively given the real-time nature of the adaptation process.

Domain generalization for prostate segmentation in transrectal ultrasound images: A multi-center study

Prostate biopsy and image-guided treatment procedures are often performed under the guidance of ultrasound fused with magnetic resonance images (MRI). Accurate image fusion relies on accurate segmentation of the prostate on ultrasound images. Yet, the reduced signal-to-noise ratio and artifacts (e.g., speckle and shadowing) in ultrasound images limit the performance of automated prostate segmentation techniques and generalizing these methods to new image domains is inherently difficult. In this study, we address these challenges by introducing a novel 2.5D deep neural network for prostate segmentation on ultrasound images. Our approach addresses the limitations of transfer learning and finetuning methods (i.e., drop in performance on the original training data when the model weights are updated) by combining a supervised domain adaptation technique and a knowledge distillation loss. The knowledge distillation loss allows the preservation of previously learned knowledge and reduces the performance drop after model finetuning on new datasets. Furthermore, our approach relies on an attention module that considers model feature positioning information to improve the segmentation accuracy. We trained our model on 764 subjects from one institution and finetuned our model using only ten subjects from subsequent institutions. We analyzed the performance of our method on three large datasets encompassing 2067 subjects from three different institutions. Our method achieved an average Dice Similarity Coefficient (Dice) of 94.0±0.03 and Hausdorff Distance (HD95) of 2.28 mm in an independent set of subjects from the first institution. Moreover, our model generalized well in the studies from the other two institutions (Dice: 91.0±0.03; HD95: 3.7 mm and Dice: 82.0±0.03; HD95: 7.1 mm). We introduced an approach that successfully segmented the prostate on ultrasound images in a multi-center study, suggesting its clinical potential to facilitate the accurate fusion of ultrasound and MRI images to drive biopsy and image-guided treatments.

The Accuracy and Radiomics Feature Effects of Multiple U-net-Based Automatic Segmentation Models for Transvaginal Ultrasound Images of Cervical Cancer

Ultrasound (US) imaging has been recognized and widely used as a screening and diagnostic imaging modality for cervical cancer all over the world. However, few studies have investigated the U-net-based automatic segmentation models for cervical cancer on US images and investigated the effects of automatic segmentations on radiomics features. A total of 1102 transvaginal US images from 796 cervical cancer patients were collected and randomly divided into training (800), validation (100) and test sets (202), respectively, in this study. Four U-net models (U-net, U-net with ResNet, context encoder network (CE-net), and Attention U-net) were adapted to segment the target of cervical cancer automatically on these US images. Radiomics features were extracted and evaluated from both manually and automatically segmented area. The mean Dice similarity coefficient (DSC) of U-net, Attention U-net, CE-net, and U-net with ResNet were 0.88, 0.89, 0.88, and 0.90, respectively. The average Pearson coefficients for the evaluation of the reliability of US image-based radiomics were 0.94, 0.96, 0.94, and 0.95 for U-net, U-net with ResNet, Attention U-net, and CE-net, respectively, in their comparison with manual segmentation. The reproducibility of the radiomics parameters evaluated by intraclass correlation coefficients (ICC) showed robustness of automatic segmentation with an average ICC coefficient of 0.99. In conclusion, high accuracy of U-net-based automatic segmentations was achieved in delineating the target area of cervical cancer US images. It is feasible and reliable for further radiomics studies with features extracted from automatic segmented target areas.

Effect of dataset size, image quality, and image type on deep learning-based automatic prostate segmentation in 3D ultrasound

Three-dimensional (3D) transrectal ultrasound (TRUS) is utilized in prostate cancer diagnosis and treatment, necessitating time-consuming manual prostate segmentation. We have previously developed an automatic 3D prostate segmentation algorithm involving deep learning prediction on radially sampled 2D images followed by 3D reconstruction, trained on a large, clinically diverse dataset with variable image quality. As large clinical datasets are rare, widespread adoption of automatic segmentation could be facilitated with efficient 2D-based approaches and the development of an image quality grading method. The complete training dataset of 6761 2D images, resliced from 206 3D TRUS volumes acquired using end-fire and side-fire acquisition methods, was split to train two separate networks using either end-fire or side-fire images. Split datasets were reduced to 1000, 500, 250, and 100 2D images. For deep learning prediction, modified U-Net and U-Net++ architectures were implemented and compared using an unseen test dataset of 40 3D TRUS volumes. A 3D TRUS image quality grading scale with three factors (acquisition quality, artifact severity, and boundary visibility) was developed to assess the impact on segmentation performance. For the complete training dataset, U-Net and U-Net++ networks demonstrated equivalent performance, but when trained using split end-fire/side-fire datasets, U-Net++ significantly outperformed the U-Net. Compared to the complete training datasets, U-Net++ trained using reduced-size end-fire and side-fire datasets demonstrated equivalent performance down to 500 training images. For this dataset, image quality had no impact on segmentation performance for end-fire images but did have a significant effect for side-fire images, with boundary visibility having the largest impact. Our algorithm provided fast (<1.5 s) and accurate 3D segmentations across clinically diverse images, demonstrating generalizability and efficiency when employed on smaller datasets, supporting the potential for widespread use, even when data is scarce. The development of an image quality grading scale provides a quantitative tool for assessing segmentation performance.

A review of artificial intelligence in prostate cancer detection on imaging

A multitude of studies have explored the role of artificial intelligence (AI) in providing diagnostic support to radiologists, pathologists, and urologists in prostate cancer detection, risk-stratification, and management. This review provides a comprehensive overview of relevant literature regarding the use of AI models in (1) detecting prostate cancer on radiology images (magnetic resonance and ultrasound imaging), (2) detecting prostate cancer on histopathology images of prostate biopsy tissue, and (3) assisting in supporting tasks for prostate cancer detection (prostate gland segmentation, MRI-histopathology registration, MRI-ultrasound registration). We discuss both the potential of these AI models to assist in the clinical workflow of prostate cancer diagnosis, as well as the current limitations including variability in training data sets, algorithms, and evaluation criteria. We also discuss ongoing challenges and what is needed to bridge the gap between academic research on AI for prostate cancer and commercial solutions that improve routine clinical care.

Real-time multimodal image registration with partial intraoperative point-set data

We present Free Point Transformer (FPT) - a deep neural network architecture for non-rigid point-set registration. Consisting of two modules, a global feature extraction module and a point transformation module, FPT does not assume explicit constraints based on point vicinity, thereby overcoming a common requirement of previous learning-based point-set registration methods. FPT is designed to accept unordered and unstructured point-sets with a variable number of points and uses a "model-free" approach without heuristic constraints. Training FPT is flexible and involves minimizing an intuitive unsupervised loss function, but supervised, semi-supervised, and partially- or weakly-supervised training are also supported. This flexibility makes FPT amenable to multimodal image registration problems where the ground-truth deformations are difficult or impossible to measure. In this paper, we demonstrate the application of FPT to non-rigid registration of prostate magnetic resonance (MR) imaging and sparsely-sampled transrectal ultrasound (TRUS) images. The registration errors were 4.71 mm and 4.81 mm for complete TRUS imaging and sparsely-sampled TRUS imaging, respectively. The results indicate superior accuracy to the alternative rigid and non-rigid registration algorithms tested and substantially lower computation time. The rapid inference possible with FPT makes it particularly suitable for applications where real-time registration is beneficial.

Robust and Accurate Mandible Segmentation on Dental CBCT Scans Affected by Metal Artifacts Using a Prior Shape Model

Accurate mandible segmentation is significant in the field of maxillofacial surgery to guide clinical diagnosis and treatment and develop appropriate surgical plans. In particular, cone-beam computed tomography (CBCT) images with metal parts, such as those used in oral and maxillofacial surgery (OMFS), often have susceptibilities when metal artifacts are present such as weak and blurred boundaries caused by a high-attenuation material and a low radiation dose in image acquisition. To overcome this problem, this paper proposes a novel deep learning-based approach (SASeg) for automated mandible segmentation that perceives overall mandible anatomical knowledge. SASeg utilizes a prior shape feature extractor (PSFE) module based on a mean mandible shape, and recurrent connections maintain the continuity structure of the mandible. The effectiveness of the proposed network is substantiated on a dental CBCT dataset from orthodontic treatment containing 59 patients. The experiments show that the proposed SASeg can be easily used to improve the prediction accuracy in a dental CBCT dataset corrupted by metal artifacts. In addition, the experimental results on the PDDCA dataset demonstrate that, compared with the state-of-the-art mandible segmentation models, our proposed SASeg can achieve better segmentation performance.

Multiple U-Net-Based Automatic Segmentations and Radiomics Feature Stability on Ultrasound Images for Patients With Ovarian Cancer

Few studies have reported the reproducibility and stability of ultrasound (US) images based radiomics features obtained from automatic segmentation in oncology. The purpose of this study is to study the accuracy of automatic segmentation algorithms based on multiple U-net models and their effects on radiomics features from US images for patients with ovarian cancer. A total of 469 US images from 127 patients were collected and randomly divided into three groups: training sets (353 images), validation sets (23 images), and test sets (93 images) for automatic segmentation models building. Manual segmentation of target volumes was delineated as ground truth. Automatic segmentations were conducted with U-net, U-net++, U-net with Resnet as the backbone (U-net with Resnet), and CE-Net. A python 3.7.0 and package Pyradiomics 2.2.0 were used to extract radiomic features from the segmented target volumes. The accuracy of automatic segmentations was evaluated by Jaccard similarity coefficient (JSC), dice similarity coefficient (DSC), and average surface distance (ASD). The reliability of radiomics features were evaluated by Pearson correlation and intraclass correlation coefficients (ICC). CE-Net and U-net with Resnet outperformed U-net and U-net++ in accuracy performance by achieving a DSC, JSC, and ASD of 0.87, 0.79, 8.54, and 0.86, 0.78, 10.00, respectively. A total of 97 features were extracted from the delineated target volumes. The average Pearson correlation was 0.86 (95% CI, 0.83–0.89), 0.87 (95% CI, 0.84–0.90), 0.88 (95% CI, 0.86–0.91), and 0.90 (95% CI, 0.88–0.92) for U-net++, U-net, U-net with Resnet, and CE-Net, respectively. The average ICC was 0.84 (95% CI, 0.81–0.87), 0.85 (95% CI, 0.82–0.88), 0.88 (95% CI, 0.85–0.90), and 0.89 (95% CI, 0.86–0.91) for U-net++, U-net, U-net with Resnet, and CE-Net, respectively. CE-Net based segmentation achieved the best radiomics reliability. In conclusion, U-net based automatic segmentation was accurate enough to delineate the target volumes on US images for patients with ovarian cancer. Radiomics features extracted from automatic segmented targets showed good reproducibility and for reliability further radiomics investigations.

Evaluation of multislice inputs to convolutional neural networks for medical image segmentation

PURPOSE

When using convolutional neural networks (CNNs) for segmentation of organs and lesions in medical images, the conventional approach is to work with inputs and outputs either as single slice [two-dimensional (2D)] or whole volumes [three-dimensional (3D)]. One common alternative, in this study denoted as pseudo-3D, is to use a stack of adjacent slices as input and produce a prediction for at least the central slice. This approach gives the network the possibility to capture 3D spatial information, with only a minor additional computational cost.

METHODS

In this study, we systematically evaluate the segmentation performance and computational costs of this pseudo-3D approach as a function of the number of input slices, and compare the results to conventional end-to-end 2D and 3D CNNs, and to triplanar orthogonal 2D CNNs. The standard pseudo-3D method regards the neighboring slices as multiple input image channels. We additionally design and evaluate a novel, simple approach where the input stack is a volumetric input that is repeatably convolved in 3D to obtain a 2D feature map. This 2D map is in turn fed into a standard 2D network. We conducted experiments using two different CNN backbone architectures and on eight diverse data sets covering different anatomical regions, imaging modalities, and segmentation tasks.

RESULTS

We found that while both pseudo-3D methods can process a large number of slices at once and still be computationally much more efficient than fully 3D CNNs, a significant improvement over a regular 2D CNN was only observed with two of the eight data sets. triplanar networks had the poorest performance of all the evaluated models. An analysis of the structural properties of the segmentation masks revealed no relations to the segmentation performance with respect to the number of input slices. A post hoc rank sum test which combined all metrics and data sets yielded that only our newly proposed pseudo-3D method with an input size of 13 slices outperformed almost all methods.

CONCLUSION

In the general case, multislice inputs appear not to improve segmentation results over using 2D or 3D CNNs. For the particular case of 13 input slices, the proposed novel pseudo-3D method does appear to have a slight advantage across all data sets compared to all other methods evaluated in this work.

FastSurfer - A fast and accurate deep learning based neuroimaging pipeline

Traditional neuroimage analysis pipelines involve computationally intensive, time-consuming optimization steps, and thus, do not scale well to large cohort studies with thousands or tens of thousands of individuals. In this work we propose a fast and accurate deep learning based neuroimaging pipeline for the automated processing of structural human brain MRI scans, replicating FreeSurfer's anatomical segmentation including surface reconstruction and cortical parcellation. To this end, we introduce an advanced deep learning architecture capable of whole-brain segmentation into 95 classes. The network architecture incorporates local and global competition via competitive dense blocks and competitive skip pathways, as well as multi-slice information aggregation that specifically tailor network performance towards accurate segmentation of both cortical and subcortical structures. Further, we perform fast cortical surface reconstruction and thickness analysis by introducing a spectral spherical embedding and by directly mapping the cortical labels from the image to the surface. This approach provides a full FreeSurfer alternative for volumetric analysis (in under 1 ​min) and surface-based thickness analysis (within only around 1 ​h runtime). For sustainability of this approach we perform extensive validation: we assert high segmentation accuracy on several unseen datasets, measure generalizability and demonstrate increased test-retest reliability, and high sensitivity to group differences in dementia.

Current status of deep learning applications in abdominal ultrasonography

Deep learning is one of the most popular artificial intelligence techniques used in the medical field. Although it is at an early stage compared to deep learning analyses of computed tomography or magnetic resonance imaging, studies applying deep learning to ultrasound imaging have been actively conducted. This review analyzes recent studies that applied deep learning to ultrasound imaging of various abdominal organs and explains the challenges encountered in these applications.

A deep learning method for real-time intraoperative US image segmentation in prostate brachytherapy

PURPOSE

This paper addresses the detection of the clinical target volume (CTV) in transrectal ultrasound (TRUS) image-guided intraoperative for permanent prostate brachytherapy. Developing a robust and automatic method to detect the CTV on intraoperative TRUS images is clinically important to have faster and reproducible interventions that can benefit both the clinical workflow and patient health.

METHODS

We present a multi-task deep learning method for an automatic prostate CTV boundary detection in intraoperative TRUS images by leveraging both the low-level and high-level (prior shape) information. Our method includes a channel-wise feature calibration strategy for low-level feature extraction and learning-based prior knowledge modeling for prostate CTV shape reconstruction. It employs CTV shape reconstruction from automatically sampled boundary surface coordinates (pseudo-landmarks) to detect the low-contrast and noisy regions across the prostate boundary, while being less biased from shadowing, inherent speckles, and artifact signals from the needle and implanted radioactive seeds.

RESULTS

The proposed method was evaluated on a clinical database of 145 patients who underwent permanent prostate brachytherapy under TRUS guidance. Our method achieved a mean accuracy of [Formula: see text] and a mean surface distance error of [Formula: see text]. Extensive ablation and comparison studies show that our method outperformed previous deep learning-based methods by more than 7% for the Dice similarity coefficient and 6.9 mm reduced 3D Hausdorff distance error.

CONCLUSION

Our study demonstrates the potential of shape model-based deep learning methods for an efficient and accurate CTV segmentation in an ultrasound-guided intervention. Moreover, learning both low-level features and prior shape knowledge with channel-wise feature calibration can significantly improve the performance of deep learning methods in medical image segmentation.

MRI Cross-Modality Image-to-Image Translation

We present a cross-modality generation framework that learns to generate translated modalities from given modalities in MR images. Our proposed method performs Image Modality Translation (abbreviated as IMT) by means of a deep learning model that leverages conditional generative adversarial networks (cGANs). Our framework jointly exploits the low-level features (pixel-wise information) and high-level representations (e.g. brain tumors, brain structure like gray matter, etc.) between cross modalities which are important for resolving the challenging complexity in brain structures. Our framework can serve as an auxiliary method in medical use and has great application potential. Based on our proposed framework, we first propose a method for cross-modality registration by fusing the deformation fields to adopt the cross-modality information from translated modalities. Second, we propose an approach for MRI segmentation, translated multichannel segmentation (TMS), where given modalities, along with translated modalities, are segmented by fully convolutional networks (FCN) in a multichannel manner. Both of these two methods successfully adopt the cross-modality information to improve the performance without adding any extra data. Experiments demonstrate that our proposed framework advances the state-of-the-art on five brain MRI datasets. We also observe encouraging results in cross-modality registration and segmentation on some widely adopted brain datasets. Overall, our work can serve as an auxiliary method in medical use and be applied to various tasks in medical fields.

Automatic segmentation of prostate MRI using convolutional neural networks: Investigating the impact of network architecture on the accuracy of volume measurement and MRI-ultrasound registration

Convolutional neural networks (CNNs) have recently led to significant advances in automatic segmentations of anatomical structures in medical images, and a wide variety of network architectures are now available to the research community. For applications such as segmentation of the prostate in magnetic resonance images (MRI), the results of the PROMISE12 online algorithm evaluation platform have demonstrated differences between the best-performing segmentation algorithms in terms of numerical accuracy using standard metrics such as the Dice score and boundary distance. These small differences in the segmented regions/boundaries outputted by different algorithms may potentially have an unsubstantial impact on the results of downstream image analysis tasks, such as estimating organ volume and multimodal image registration, which inform clinical decisions. This impact has not been previously investigated. In this work, we quantified the accuracy of six different CNNs in segmenting the prostate in 3D patient T2-weighted MRI scans and compared the accuracy of organ volume estimation and MRI-ultrasound (US) registration errors using the prostate segmentations produced by different networks. Networks were trained and tested using a set of 232 patient MRIs with labels provided by experienced clinicians. A statistically significant difference was found among the Dice scores and boundary distances produced by these networks in a non-parametric analysis of variance (p < 0.001 and p < 0.001, respectively), where the following multiple comparison tests revealed that the statistically significant difference in segmentation errors were caused by at least one tested network. Gland volume errors (GVEs) and target registration errors (TREs) were then estimated using the CNN-generated segmentations. Interestingly, there was no statistical difference found in either GVEs or TREs among different networks, (p = 0.34 and p = 0.26, respectively). This result provides a real-world example that these networks with different segmentation performances may potentially provide indistinguishably adequate registration accuracies to assist prostate cancer imaging applications. We conclude by recommending that the differences in the accuracy of downstream image analysis tasks that make use of data output by automatic segmentation methods, such as CNNs, within a clinical pipeline should be taken into account when selecting between different network architectures, in addition to reporting the segmentation accuracy.

Deep Attentive Features for Prostate Segmentation in 3D Transrectal Ultrasound

Automatic prostate segmentation in transrectal ultrasound (TRUS) images is of essential importance for image-guided prostate interventions and treatment planning. However, developing such automatic solutions remains very challenging due to the missing/ambiguous boundary and inhomogeneous intensity distribution of the prostate in TRUS, as well as the large variability in prostate shapes. This paper develops a novel 3D deep neural network equipped with attention modules for better prostate segmentation in TRUS by fully exploiting the complementary information encoded in different layers of the convolutional neural network (CNN). Our attention module utilizes the attention mechanism to selectively leverage the multi-level features integrated from different layers to refine the features at each individual layer, suppressing the non-prostate noise at shallow layers of the CNN and increasing more prostate details into features at deep layers. Experimental results on challenging 3D TRUS volumes show that our method attains satisfactory segmentation performance. The proposed attention mechanism is a general strategy to aggregate multi-level deep features and has the potential to be used for other medical image segmentation tasks. The code is publicly available at https://github.com/wulalago/DAF3D.

Accurate and robust deep learning-based segmentation of the prostate clinical target volume in ultrasound images

The goal of this work was to develop a method for accurate and robust automatic segmentation of the prostate clinical target volume in transrectal ultrasound (TRUS) images for brachytherapy. These images can be difficult to segment because of weak or insufficient landmarks or strong artifacts. We devise a method, based on convolutional neural networks (CNNs), that produces accurate segmentations on easy and difficult images alike. We propose two strategies to achieve improved segmentation accuracy on difficult images. First, for CNN training we adopt an adaptive sampling strategy, whereby the training process is encouraged to pay more attention to images that are difficult to segment. Secondly, we train a CNN ensemble and use the disagreement among this ensemble to identify uncertain segmentations and to estimate a segmentation uncertainty map. We improve uncertain segmentations by utilizing the prior shape information in the form of a statistical shape model. Our method achieves Hausdorff distance of 2.7 ± 2.3 mm and Dice score of 93.9 ± 3.5%. Comparisons with several competing methods show that our method achieves significantly better results and reduces the likelihood of committing large segmentation errors. Furthermore, our experiments show that our approach to estimating segmentation uncertainty is better than or on par with recent methods for estimation of prediction uncertainty in deep learning models. Our study demonstrates that estimation of model uncertainty and use of prior shape information can significantly improve the performance of CNN-based medical image segmentation methods, especially on difficult images.

Weakly-supervised convolutional neural networks for multimodal image registration

One of the fundamental challenges in supervised learning for multimodal image registration is the lack of ground-truth for voxel-level spatial correspondence. This work describes a method to infer voxel-level transformation from higher-level correspondence information contained in anatomical labels. We argue that such labels are more reliable and practical to obtain for reference sets of image pairs than voxel-level correspondence. Typical anatomical labels of interest may include solid organs, vessels, ducts, structure boundaries and other subject-specific ad hoc landmarks. The proposed end-to-end convolutional neural network approach aims to predict displacement fields to align multiple labelled corresponding structures for individual image pairs during the training, while only unlabelled image pairs are used as the network input for inference. We highlight the versatility of the proposed strategy, for training, utilising diverse types of anatomical labels, which need not to be identifiable over all training image pairs. At inference, the resulting 3D deformable image registration algorithm runs in real-time and is fully-automated without requiring any anatomical labels or initialisation. Several network architecture variants are compared for registering T2-weighted magnetic resonance images and 3D transrectal ultrasound images from prostate cancer patients. A median target registration error of 3.6 mm on landmark centroids and a median Dice of 0.87 on prostate glands are achieved from cross-validation experiments, in which 108 pairs of multimodal images from 76 patients were tested with high-quality anatomical labels.