We present our novel deep multi-task learning method for medical image segmentation. Existing multi-task methods demand ground truth annotations for both the primary and auxiliary tasks. Contrary to it, we propose to generate the pseudo-labels of an auxiliary task in an unsupervised manner. To generate the pseudo-labels, we leverage Histogram of Oriented Gradients (HOGs), one of the most widely used and powerful hand-crafted features for detection. Together with the ground truth semantic segmentation masks for the primary task and pseudo-labels for the auxiliary task, we learn the parameters of the deep network to minimise the loss of both the primary task and the auxiliary task jointly. We employed our method on two powerful and widely used semantic segmentation networks: UNet and U2Net to train in a multi-task setup. To validate our hypothesis, we performed experiments on two different medical image segmentation data sets. From the extensive quantitative and qualitative results, we observe that our method consistently improves the performance compared to the counter-part method. Moreover, our method is the winner of FetReg Endovis Sub-challenge on Semantic Segmentation organised in conjunction with MICCAI 2021.
The segmentation of diseases is a popular topic explored by researchers in the field of machine learning. Brain tumors are extremely dangerous and require the utmost precision to segment for a successful surgery. Patients with tumors usually take 4 MRI scans, T1, T1gd, T2, and FLAIR, which are then sent to radiologists to segment and analyze for possible future surgery. To create a second segmentation, it would be beneficial to both radiologists and patients in being more confident in their conclusions. We propose using a method performed by radiologists called image segmentation and applying it to machine learning models to prove a better segmentation. Using Mask R-CNN, its ResNet backbone being pre-trained on the RSNA pneumonia detection challenge dataset, we can train a model on the Brats2020 Brain Tumor dataset. Center for Biomedical Image Computing & Analytics provides MRI data on patients with and without brain tumors and the corresponding segmentations. We can see how well the method of image subtraction works by comparing it to models without image subtraction through DICE coefficient (F1 score), recall, and precision on the untouched test set. Our model performed with a DICE coefficient of 0.75 in comparison to 0.69 without image subtraction. To further emphasize the usefulness of image subtraction, we compare our final model to current state-of-the-art models to segment tumors from MRI scans.
Clustering can be defined as the process of assembling objects into a number of groups whose elements are similar to each other in some manner. As a technique that is used in many domains, such as face clustering, plant categorization, image segmentation, document classification, clustering is considered one of the most important unsupervised learning problems. Scientists have surveyed this problem for years and developed different techniques that can solve it, such as k-means clustering. We analyze one of these techniques: a powerful clustering algorithm called Sparse Subspace Clustering. We demonstrate several experiments using this method and then introduce a new approach that can reduce the computational time required to perform sparse subspace clustering.
Vision Transformers (ViT)s have recently become popular due to their outstanding modeling capabilities, in particular for capturing long-range information, and scalability to dataset and model sizes which has led to state-of-the-art performance in various computer vision and medical image analysis tasks. In this work, we introduce a unified framework consisting of two architectures, dubbed UNetFormer, with a 3D Swin Transformer-based encoder and Convolutional Neural Network (CNN) and transformer-based decoders. In the proposed model, the encoder is linked to the decoder via skip connections at five different resolutions with deep supervision. The design of proposed architecture allows for meeting a wide range of trade-off requirements between accuracy and computational cost. In addition, we present a methodology for self-supervised pre-training of the encoder backbone via learning to predict randomly masked volumetric tokens using contextual information of visible tokens. We pre-train our framework on a cohort of $5050$ CT images, gathered from publicly available CT datasets, and present a systematic investigation of various components such as masking ratio and patch size that affect the representation learning capability and performance of downstream tasks. We validate the effectiveness of our pre-training approach by fine-tuning and testing our model on liver and liver tumor segmentation task using the Medical Segmentation Decathlon (MSD) dataset and achieve state-of-the-art performance in terms of various segmentation metrics. To demonstrate its generalizability, we train and test the model on BraTS 21 dataset for brain tumor segmentation using MRI images and outperform other methods in terms of Dice score. Code: https://github.com/Project-MONAI/research-contributions
This paper shows the application of autonomous Crater Detection using the U-Net, a Fully-Convolutional Neural Network, on Ceres. The U-Net is trained on optical images of the Moon Global Morphology Mosaic based on data collected by the LRO and manual crater catalogues. The Moon-trained network will be tested on Dawn optical images of Ceres: this task is accomplished by means of a Transfer Learning (TL) approach. The trained model has been fine-tuned using 100, 500 and 1000 additional images of Ceres. The test performance was measured on 350 never before seen images, reaching a testing accuracy of 96.24%, 96.95% and 97.19%, respectively. This means that despite the intrinsic differences between the Moon and Ceres, TL works with encouraging results. The output of the U-Net contains predicted craters: it will be post-processed applying global thresholding for image binarization and a template matching algorithm to extract craters positions and radii in the pixel space. Post-processed craters will be counted and compared to the ground truth data in order to compute image segmentation metrics: precision, recall and F1 score. These indices will be computed, and their effect will be discussed for tasks such as automated crater cataloguing and optical navigation.
Real-time holistic scene understanding would allow machines to interpret their surrounding in a much more detailed manner than is currently possible. While panoptic image segmentation methods have brought image segmentation closer to this goal, this information has to be described relative to the 3D environment for the machine to be able to utilise it effectively. In this paper, we investigate methods for sequentially reconstructing static environments from panoptic image segmentations in 3D. We specifically target real-time operation: the algorithm must process data strictly online and be able to run at relatively fast frame rates. Additionally, the method should be scalable for environments large enough for practical applications. By applying a simple but powerful data-association algorithm, we outperform earlier similar works when operating purely online. Our method is also capable of reaching frame-rates high enough for real-time applications and is scalable to larger environments as well. Source code and further demonstrations are released to the public at: \url{https://tutvision.github.io/Online-Panoptic-3D/}
Alzheimer's disease (AD) is one of the most common public health issues the world is facing today. This disease has a high prevalence primarily in the elderly accompanying memory loss and cognitive decline. AD detection is a challenging task which many authors have developed numerous computerized automatic diagnosis systems utilizing neuroimaging and other clinical data. MRI scans provide high-intensity visible features, making these scans the most widely used brain imaging technique. In recent years deep learning has achieved leading success in medical image analysis. But a relatively little investigation has been done to apply deep learning techniques for the brain MRI classification. This paper explores the construction of several deep learning architectures evaluated on brain MRI images and segmented images. The idea behind segmented images investigates the influence of image segmentation step on deep learning classification. The image processing presented a pipeline consisting of pre-processing to enhance the MRI scans and post-processing consisting of a segmentation method for segmenting the brain tissues. The results show that the processed images achieved a better accuracy in the binary classification of AD vs. CN (Cognitively Normal) across four different architectures. ResNet architecture resulted in the highest prediction accuracy amongst the other architectures (90.83% for the original brain images and 93.50% for the processed images).
Recent advances in deep learning-based medical image segmentation studies achieve nearly human-level performance when in fully supervised condition. However, acquiring pixel-level expert annotations is extremely expensive and laborious in medical imaging fields. Unsupervised domain adaptation can alleviate this problem, which makes it possible to use annotated data in one imaging modality to train a network that can successfully perform segmentation on target imaging modality with no labels. In this work, we propose a self-training based unsupervised domain adaptation framework for 3D medical image segmentation named COSMOS and validate it with automatic segmentation of Vestibular Schwannoma (VS) and cochlea on high-resolution T2 Magnetic Resonance Images (MRI). Our target-aware contrast conversion network translates source domain annotated T1 MRI to pseudo T2 MRI to enable segmentation training on target domain, while preserving important anatomical features of interest in the converted images. Iterative self-training is followed to incorporate unlabeled data to training and incrementally improve the quality of pseudo-labels, thereby leading to improved performance of segmentation. COSMOS won the 1\textsuperscript{st} place in the Cross-Modality Domain Adaptation (crossMoDA) challenge held in conjunction with the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2021). It achieves mean Dice score and Average Symmetric Surface Distance of 0.871(0.063) and 0.437(0.270) for VS, and 0.842(0.020) and 0.152(0.030) for cochlea.
Referring image segmentation is an advanced semantic segmentation task where target is not a predefined class but is described in natural language. Most of existing methods for this task rely heavily on convolutional neural networks, which however have trouble capturing long-range dependencies between entities in the language expression and are not flexible enough for modeling interactions between the two different modalities. To address these issues, we present the first convolution-free model for referring image segmentation using transformers, dubbed ReSTR. Since it extracts features of both modalities through transformer encoders, it can capture long-range dependencies between entities within each modality. Also, ReSTR fuses features of the two modalities by a self-attention encoder, which enables flexible and adaptive interactions between the two modalities in the fusion process. The fused features are fed to a segmentation module, which works adaptively according to the image and language expression in hand. ReSTR is evaluated and compared with previous work on all public benchmarks, where it outperforms all existing models.
We propose a new problem formulation and a corresponding evaluation framework to advance research on unsupervised domain adaptation for semantic image segmentation. The overall goal is fostering the development of adaptive learning systems that will continuously learn, without supervision, in ever-changing environments. Typical protocols that study adaptation algorithms for segmentation models are limited to few domains, adaptation happens offline, and human intervention is generally required, at least to annotate data for hyper-parameter tuning. We argue that such constraints are incompatible with algorithms that can continuously adapt to different real-world situations. To address this, we propose a protocol where models need to learn online, from sequences of temporally correlated images, requiring continuous, frame-by-frame adaptation. We accompany this new protocol with a variety of baselines to tackle the proposed formulation, as well as an extensive analysis of their behaviors, which can serve as a starting point for future research.
We propose a test-time adaptation method for cross-domain image segmentation. Our method is simple: Given a new unseen instance at test time, we adapt a pre-trained model by conducting instance-specific BatchNorm (statistics) calibration. Our approach has two core components. First, we replace the manually designed BatchNorm calibration rule with a learnable module. Second, we leverage strong data augmentation to simulate random domain shifts for learning the calibration rule. In contrast to existing domain adaptation methods, our method does not require accessing the target domain data at training time or conducting computationally expensive test-time model training/optimization. Equipping our method with models trained by standard recipes achieves significant improvement, comparing favorably with several state-of-the-art domain generalization and one-shot unsupervised domain adaptation approaches. Combining our method with the domain generalization methods further improves performance, reaching a new state of the art.
In this research, we propose a tiny image segmentation model, L^3U-net, that works on low-resource edge devices in real-time. We introduce a data folding technique that reduces inference latency by leveraging the parallel convolutional layer processing capability of the CNN accelerators. We also deploy the proposed model to such a device, MAX78000, and the results show that L^3U-net achieves more than 90% accuracy over two different segmentation datasets with 10 fps.
Dice similarity coefficient (DSC) and Hausdorff distance (HD) are widely used for evaluating medical image segmentation. They have also been criticised, when reported alone, for their unclear or even misleading clinical interpretation. DSCs may also differ substantially from HDs, due to boundary smoothness or multiple regions of interest (ROIs) within a subject. More importantly, either metric can also have a nonlinear, non-monotonic relationship with outcomes based on Type 1 and 2 errors, designed for specific clinical decisions that use the resulting segmentation. Whilst cases causing disagreement between these metrics are not difficult to postulate. This work first proposes a new asymmetric detection metric, adapting those used in object detection, for planning prostate cancer procedures. The lesion-level metrics is then compared with the voxel-level DSC and HD, whereas a 3D UNet is used for segmenting lesions from multiparametric MR (mpMR) images. Based on experimental results we report pairwise agreement and correlation 1) between DSC and HD, and 2) between voxel-level DSC and recall-controlled precision at lesion-level, with Cohen's [0.49, 0.61] and Pearson's [0.66, 0.76] (p-values}<0.001) at varying cut-offs. However, the differences in false-positives and false-negatives, between the actual errors and the perceived counterparts if DSC is used, can be as high as 152 and 154, respectively, out of the 357 test set lesions. We therefore carefully conclude that, despite of the significant correlations, voxel-level metrics such as DSC can misrepresent lesion-level detection accuracy for evaluating localisation of multifocal prostate cancer and should be interpreted with caution.
Synthetic aperture sonar (SAS) systems produce high-resolution images of the seabed environment. Moreover, deep learning has demonstrated superior ability in finding robust features for automating imagery analysis. However, the success of deep learning is conditioned on having lots of labeled training data, but obtaining generous pixel-level annotations of SAS imagery is often practically infeasible. This challenge has thus far limited the adoption of deep learning methods for SAS segmentation. Algorithms exist to segment SAS imagery in an unsupervised manner, but they lack the benefit of state-of-the-art learning methods and the results present significant room for improvement. In view of the above, we propose a new iterative algorithm for unsupervised SAS image segmentation combining superpixel formation, deep learning, and traditional clustering methods. We call our method Iterative Deep Unsupervised Segmentation (IDUS). IDUS is an unsupervised learning framework that can be divided into four main steps: 1) A deep network estimates class assignments. 2) Low-level image features from the deep network are clustered into superpixels. 3) Superpixels are clustered into class assignments (which we call pseudo-labels) using $k$-means. 4) Resulting pseudo-labels are used for loss backpropagation of the deep network prediction. These four steps are performed iteratively until convergence. A comparison of IDUS to current state-of-the-art methods on a realistic benchmark dataset for SAS image segmentation demonstrates the benefits of our proposal even as the IDUS incurs a much lower computational burden during inference (actual labeling of a test image). Finally, we also develop a semi-supervised (SS) extension of IDUS called IDSS and demonstrate experimentally that it can further enhance performance while outperforming supervised alternatives that exploit the same labeled training imagery.
Hyperspectral images provide a rich representation of the underlying spectrum for each pixel, allowing for a pixel-wise classification/segmentation into different classes. As the acquisition of labeled training data is very time-consuming, unsupervised methods become crucial in hyperspectral image analysis. The spectral variability and noise in hyperspectral data make this task very challenging and define special requirements for such methods. Here, we present a novel unsupervised hyperspectral segmentation framework. It starts with a denoising and dimensionality reduction step by the well-established Minimum Noise Fraction (MNF) transform. Then, the Mumford-Shah (MS) segmentation functional is applied to segment the data. We equipped the MS functional with a novel robust distribution-dependent indicator function designed to handle the characteristic challenges of hyperspectral data. To optimize our objective function with respect to the parameters for which no closed form solution is available, we propose an efficient fixed point iteration scheme. Numerical experiments on four public benchmark datasets show that our method produces competitive results, which outperform three state-of-the-art methods substantially on three of these datasets.
Image segmentation is a crucial but challenging task that has many applications. In medical imaging for instance, intensity inhomogeneity and noise are common. In thigh muscle images, different muscles are closed packed together and there are often no clear boundaries between them. Intensity based segmentation models cannot separate one muscle from another. To solve such problems, in this work we present a segmentation model with adaptive spatial priors from joint registration. This model combines segmentation and registration in a unified framework to leverage their positive mutual influence. The segmentation is based on a modified Gaussian mixture model (GMM), which integrates intensity inhomogeneity and spacial smoothness. The registration plays the role of providing a shape prior. We adopt a modified sum of squared difference (SSD) fidelity term and Tikhonov regularity term for registration, and also utilize Gaussian pyramid and parametric method for robustness. The connection between segmentation and registration is guaranteed by the cross entropy metric that aims to make the segmentation map (from segmentation) and deformed atlas (from registration) as similar as possible. This joint framework is implemented within a constraint optimization framework, which leads to an efficient algorithm. We evaluate our proposed model on synthetic and thigh muscle MR images. Numerical results show the improvement as compared to segmentation and registration performed separately and other joint models.
Synthesizing a subject-specific pathology-free image from a pathological image is valuable for algorithm development and clinical practice. In recent years, several approaches based on the Generative Adversarial Network (GAN) have achieved promising results in pseudo-healthy synthesis. However, the discriminator (i.e., a classifier) in the GAN cannot accurately identify lesions and further hampers from generating admirable pseudo-healthy images. To address this problem, we present a new type of discriminator, the segmentor, to accurately locate the lesions and improve the visual quality of pseudo-healthy images. Then, we apply the generated images into medical image enhancement and utilize the enhanced results to cope with the low contrast problem existing in medical image segmentation. Furthermore, a reliable metric is proposed by utilizing two attributes of label noise to measure the health of synthetic images. Comprehensive experiments on the T2 modality of BraTS demonstrate that the proposed method substantially outperforms the state-of-the-art methods. The method achieves better performance than the existing methods with only 30\% of the training data. The effectiveness of the proposed method is also demonstrated on the LiTS and the T1 modality of BraTS. The code and the pre-trained model of this study are publicly available at https://github.com/Au3C2/Generator-Versus-Segmentor.
Solving the domain shift problem during inference is essential in medical imaging, as most deep-learning based solutions suffer from it. In practice, domain shifts are tackled by performing Unsupervised Domain Adaptation (UDA), where a model is adapted to an unlabelled target domain by leveraging the labelled source data. In medical scenarios, the data comes with huge privacy concerns making it difficult to apply standard UDA techniques. Hence, a closer clinical setting is Source-Free UDA (SFUDA), where we have access to source-trained model but not the source data during adaptation. Existing SFUDA methods rely on pseudo-label based self-training techniques to address the domain shift. However, these pseudo-labels often have high entropy due to domain shift and adapting the source model with noisy pseudo-labels leads to sub-optimal performance. To overcome this limitation, we propose a systematic two-stage approach for SFUDA comprising of target-specific adaptation followed by task-specific adaptation. In target-specific adaptation, we enhance the pseudo-label generation by minimizing high entropy regions using the proposed ensemble entropy minimization loss and a selective voting strategy. In task-specific adaptation, we exploit the enhanced pseudo-labels using a student-teacher framework to effectively learn segmentation on the target domain. We evaluate our proposed method on 2D fundus datasets and 3D MRI volumes across 7 different domain shifts where we perform better than existing UDA and SFUDA methods for medical image segmentation. Code is available at https://github.com/Vibashan/tt-sfuda.
Automatic fetal brain tissue segmentation can enhance the quantitative assessment of brain development at this critical stage. Deep learning methods represent the state of the art in medical image segmentation and have also achieved impressive results in brain segmentation. However, effective training of a deep learning model to perform this task requires a large number of training images to represent the rapid development of the transient fetal brain structures. On the other hand, manual multi-label segmentation of a large number of 3D images is prohibitive. To address this challenge, we segmented 272 training images, covering 19-39 gestational weeks, using an automatic multi-atlas segmentation strategy based on deformable registration and probabilistic atlas fusion, and manually corrected large errors in those segmentations. Since this process generated a large training dataset with noisy segmentations, we developed a novel label smoothing procedure and a loss function to train a deep learning model with smoothed noisy segmentations. Our proposed methods properly account for the uncertainty in tissue boundaries. We evaluated our method on 23 manually-segmented test images of a separate set of fetuses. Results show that our method achieves an average Dice similarity coefficient of 0.893 and 0.916 for the transient structures of younger and older fetuses, respectively. Our method generated results that were significantly more accurate than several state-of-the-art methods including nnU-Net that achieved the closest results to our method. Our trained model can serve as a valuable tool to enhance the accuracy and reproducibility of fetal brain analysis in MRI.
Recently, it has attracted more and more attentions to fuse multi-scale features for semantic image segmentation. Various works were proposed to employ progressive local or global fusion, but the feature fusions are not rich enough for modeling multi-scale context features. In this work, we focus on fusing multi-scale features from Transformer-based backbones for semantic segmentation, and propose a Feature Selective Transformer (FeSeFormer), which aggregates features from all scales (or levels) for each query feature. Specifically, we first propose a Scale-level Feature Selection (SFS) module, which can choose an informative subset from the whole multi-scale feature set for each scale, where those features that are important for the current scale (or level) are selected and the redundant are discarded. Furthermore, we propose a Full-scale Feature Fusion (FFF) module, which can adaptively fuse features of all scales for queries. Based on the proposed SFS and FFF modules, we develop a Feature Selective Transformer (FeSeFormer), and evaluate our FeSeFormer on four challenging semantic segmentation benchmarks, including PASCAL Context, ADE20K, COCO-Stuff 10K, and Cityscapes, outperforming the state-of-the-art.