Image-to-Images Translation for Multi-Task Organ Segmentation and Bone Suppression in Chest X-Ray Radiography

Brain Image Segmentation for Ultrascale Neuron Reconstruction via an Adaptive Dual-Task Learning Network

Accurate morphological reconstruction of neurons in whole brain images is critical for brain science research. However, due to the wide range of whole brain imaging, uneven staining, and optical system fluctuations, there are significant differences in image properties between different regions of the ultrascale brain image, such as dramatically varying voxel intensities and inhomogeneous distribution of background noise, posing an enormous challenge to neuron reconstruction from whole brain images. In this paper, we propose an adaptive dual-task learning network (ADTL-Net) to quickly and accurately extract neuronal structures from ultrascale brain images. Specifically, this framework includes an External Features Classifier (EFC) and a Parameter Adaptive Segmentation Decoder (PASD), which share the same Multi-Scale Feature Encoder (MSFE). MSFE introduces an attention module named Channel Space Fusion Module (CSFM) to extract structure and intensity distribution features of neurons at different scales for addressing the problem of anisotropy in 3D space. Then, EFC is designed to classify these feature maps based on external features, such as foreground intensity distributions and image smoothness, and select specific PASD parameters to decode them of different classes to obtain accurate segmentation results. PASD contains multiple sets of parameters trained by different representative complex signal-to-noise distribution image blocks to handle various images more robustly. Experimental results prove that compared with other advanced segmentation methods for neuron reconstruction, the proposed method achieves state-of-the-art results in the task of neuron reconstruction from ultrascale brain images, with an improvement of about 49% in speed and 12% in F1 score.

Cross-domain colorization of unpaired infrared images through contrastive learning guided by color feature selection attention

In challenging lighting conditions, infrared detectors have become vital tools for enhancing visual perception, overcoming the limitations of visible cameras. However, inherent imaging principles and manufacturing constraints confine infrared imaging systems to grayscale, significantly impacting their utility. In comparison to visible imagery, infrared images lack detailed semantic information, color representation, and suffer from reduced contrast. While existing infrared image colorization techniques have made significant progress in improving color quality, challenges such as erroneous semantic color prediction and blurred depiction of fine details persist. Acquiring paired color images corresponding to real-world infrared scenarios poses substantial difficulties, exacerbating challenges in cross-domain colorization of infrared images. To address these critical issues, this paper introduces an innovative approach utilizing contrastive learning for unsupervised cross-domain mapping between unpaired infrared and visible color images. Additionally, we introduce a color feature selection attention module guiding rational infrared image coloring. The proposed method employs the Residual Fusion Attention Network (RFANet) as a generator, enhancing the encoder's ability to represent color and structural features. Furthermore, to ensure structural content consistency and enhance overall color style matching accuracy, we design a comprehensive joint global loss function integrating both detailed content and color style. Experimental evaluations on publicly available datasets demonstrate the superior performance of the proposed unsupervised cross-domain colorization method for infrared images compared to previous approaches.

Bone mineral density estimation from a plain X-ray image by learning decomposition into projections of bone-segmented computed tomography

Osteoporosis is a prevalent bone disease that causes fractures in fragile bones, leading to a decline in daily living activities. Dual-energy X-ray absorptiometry (DXA) and quantitative computed tomography (QCT) are highly accurate for diagnosing osteoporosis; however, these modalities require special equipment and scan protocols. To frequently monitor bone health, low-cost, low-dose, and ubiquitously available diagnostic methods are highly anticipated. In this study, we aim to perform bone mineral density (BMD) estimation from a plain X-ray image for opportunistic screening, which is potentially useful for early diagnosis. Existing methods have used multi-stage approaches consisting of extraction of the region of interest and simple regression to estimate BMD, which require a large amount of training data. Therefore, we propose an efficient method that learns decomposition into projections of bone-segmented QCT for BMD estimation under limited datasets. The proposed method achieved high accuracy in BMD estimation, where Pearson correlation coefficients of 0.880 and 0.920 were observed for DXA-measured BMD and QCT-measured BMD estimation tasks, respectively, and the root mean square of the coefficient of variation values were 3.27 to 3.79% for four measurements with different poses. Furthermore, we conducted extensive validation experiments, including multi-pose, uncalibrated-CT, and compression experiments toward actual application in routine clinical practice.

MuSiC-ViT: A multi-task Siamese convolutional vision transformer for differentiating change from no-change in follow-up chest radiographs

A major responsibility of radiologists in routine clinical practice is to read follow-up chest radiographs (CXRs) to identify changes in a patient's condition. Diagnosing meaningful changes in follow-up CXRs is challenging because radiologists must differentiate disease changes from natural or benign variations. Here, we suggest using a multi-task Siamese convolutional vision transformer (MuSiC-ViT) with an anatomy-matching module (AMM) to mimic the radiologist's cognitive process for differentiating baseline change from no-change. MuSiC-ViT uses the convolutional neural networks (CNNs) meet vision transformers model that combines CNN and transformer architecture. It has three major components: a Siamese network architecture, an AMM, and multi-task learning. Because the input is a pair of CXRs, a Siamese network was adopted for the encoder. The AMM is an attention module that focuses on related regions in the CXR pairs. To mimic a radiologist's cognitive process, MuSiC-ViT was trained using multi-task learning, normal/abnormal and change/no-change classification, and anatomy-matching. Among 406 K CXRs studied, 88 K change and 115 K no-change pairs were acquired for the training dataset. The internal validation dataset consisted of 1,620 pairs. To demonstrate the robustness of MuSiC-ViT, we verified the results with two other validation datasets. MuSiC-ViT respectively achieved accuracies and area under the receiver operating characteristic curves of 0.728 and 0.797 on the internal validation dataset, 0.614 and 0.784 on the first external validation dataset, and 0.745 and 0.858 on a second temporally separated validation dataset. All code is available at https://github.com/chokyungjin/MuSiC-ViT.

A Multi-Scale Context Aware Attention Model for Medical Image Segmentation

Medical image segmentation is critical for efficient diagnosis of diseases and treatment planning. In recent years, convolutional neural networks (CNN)-based methods, particularly U-Net and its variants, have achieved remarkable results on medical image segmentation tasks. However, they do not always work consistently on images with complex structures and large variations in regions of interest (ROI). This could be due to the fixed geometric structure of the receptive fields used for feature extraction and repetitive down-sampling operations that lead to information loss. To overcome these problems, the standard U-Net architecture is modified in this work by replacing the convolution block with a dilated convolution block to extract multi-scale context features with varying sizes of receptive fields, and adding a dilated inception block between the encoder and decoder paths to alleviate the problem of information recession and the semantic gap between features. Furthermore, the input of each dilated convolution block is added to the output through a squeeze and excitation unit, which alleviates the vanishing gradient problem and improves overall feature representation by re-weighting the channel-wise feature responses. The original inception block is modified by reducing the size of the spatial filter and introducing dilated convolution to obtain a larger receptive field. The proposed network was validated on three challenging medical image segmentation tasks with varying size ROIs: lung segmentation on chest X-ray (CXR) images, skin lesion segmentation on dermoscopy images and nucleus segmentation on microscopy cell images. Improved performance compared to state-of-the-art techniques demonstrates the effectiveness and generalisability of the proposed Dilated Convolution and Inception blocks-based U-Net (DCI-UNet).

Decomposition of musculoskeletal structures from radiographs using an improved CycleGAN framework

This paper presents methods of decomposition of musculoskeletal structures from radiographs into multiple individual muscle and bone structures. While existing solutions require dual-energy scan for the training dataset and are mainly applied to structures with high-intensity contrast, such as bones, we focused on multiple superimposed muscles with subtle contrast in addition to bones. The decomposition problem is formulated as an image translation problem between (1) a real X-ray image and (2) multiple digitally reconstructed radiographs, each of which contains a single muscle or bone structure, and solved using unpaired training based on the CycleGAN framework. The training dataset was created via automatic computed tomography (CT) segmentation of muscle/bone regions and virtually projecting them with geometric parameters similar to the real X-ray images. Two additional features were incorporated into the CycleGAN framework to achieve a high-resolution and accurate decomposition: hierarchical learning and reconstruction loss with the gradient correlation similarity metric. Furthermore, we introduced a new diagnostic metric for muscle asymmetry directly measured from a plain X-ray image to validate the proposed method. Our simulation and real-image experiments using real X-ray and CT images of 475 patients with hip diseases suggested that each additional feature significantly enhanced the decomposition accuracy. The experiments also evaluated the accuracy of muscle volume ratio measurement, which suggested a potential application to muscle asymmetry assessment from an X-ray image for diagnostic and therapeutic assistance. The improved CycleGAN framework can be applied for investigating the decomposition of musculoskeletal structures from single radiographs.

Validation of an Automated Cardiothoracic Ratio Calculation for Hemodialysis Patients

Cardiomegaly is associated with poor clinical outcomes and is assessed by routine monitoring of the cardiothoracic ratio (CTR) from chest X-rays (CXRs). Judgment of the margins of the heart and lungs is subjective and may vary between different operators.

METHODS

Patients aged > 19 years in our hemodialysis unit from March 2021 to October 2021 were enrolled. The borders of the lungs and heart on CXRs were labeled by two nephrologists as the ground truth (nephrologist-defined mask). We implemented AlbuNet-34, a U-Net variant, to predict the heart and lung margins from CXR images and to automatically calculate the CTRs.

RESULTS

The coefficient of determination (R²) obtained using the neural network model was 0.96, compared with an R² of 0.90 obtained by nurse practitioners. The mean difference between the CTRs calculated by the nurse practitioners and senior nephrologists was 1.52 ± 1.46%, and that between the neural network model and the nephrologists was 0.83 ± 0.87% (p < 0.001). The mean CTR calculation duration was 85 s using the manual method and less than 2 s using the automated method (p < 0.001).

CONCLUSIONS

Our study confirmed the validity of automated CTR calculations. By achieving high accuracy and saving time, our model can be implemented in clinical practice.

Bone suppression of lateral chest x-rays with imperfect and limited dual-energy subtraction images

Bone suppression is to suppress the superimposed bone components over the soft tissues within the lung area of Chest X-ray (CXR), which is potentially useful for the subsequent lung disease diagnosis for radiologists, as well as computer-aided systems. Despite bone suppression methods for frontal CXRs being well studied, it remains challenging for lateral CXRs due to the limited and imperfect DES dataset containing paired lateral CXR and soft-tissue/bone images and more complex anatomical structures in the lateral view. In this work, we propose a bone suppression method for lateral CXRs by leveraging a two-stage distillation learning strategy and a specific data correction method. Specifically, a primary model is first trained on a real DES dataset with limited samples. The bone-suppressed results on a relatively large lateral CXR dataset produced by the primary model are improved by a designed gradient correction method. Secondly, the corrected results serve as training samples to train the distillated model. By automatically learning knowledge from both the primary model and the extra correction procedure, our distillated model is expected to promote the performance of the primary model while omitting the tedious correction procedure. We adopt an ensemble model named MsDd-MAP for the primary and distillated models, which learns the complementary information of Multi-scale and Dual-domain (i.e., intensity and gradient) and fuses them in a maximum-a-posteriori (MAP) framework. Our method is evaluated on a two-exposure lateral DES dataset consisting of 46 subjects and a lateral CXR dataset consisting of 240 subjects. The experimental results suggest that our method is superior to other competing methods regarding the quantitative evaluation metrics. Furthermore, the subjective evaluation by three experienced radiologists also indicates that the distillated model can produce more visually appealing soft-tissue images than the primary model, even comparable to real DES imaging for lateral CXRs.

X-Ray to DRR Images Translation for Efficient Multiple Objects Similarity Measures in Deformable Model 3D/2D Registration

The robustness and accuracy of the intensity-based 3D/2D registration of a 3D model on planar X-ray image(s) is related to the quality of the image correspondences between the digitally reconstructed radiographs (DRR) generated from the 3D models (varying image) and the X-ray images (fixed target). While much effort may be devoted to generating realistic DRR that are similar to real X-rays (using complex X-ray simulation, adding densities information in 3D models, etc.), significant differences still remain between DRR and real X-ray images. Differences such as the presence of adjacent or superimposed soft tissue and bony or foreign structures lead to image matching difficulties and decrease the 3D/2D registration performance. In the proposed method, the X-ray images were converted into DRR images using a GAN-based cross-modality image-to-images translation. With this added prior step of XRAY-to-DRR translation, standard similarity measures become efficient even when using simple and fast DRR projection. For both images to match, they must belong to the same image domain and essentially contain the same kind of information. The XRAY-to-DRR translation also addresses the well-known issue of registering an object in a scene composed of multiple objects by separating the superimposed or/and adjacent objects to avoid mismatching across similar structures. We applied the proposed method to the 3D/2D fine registration of vertebra deformable models to biplanar radiographs of the spine. We showed that the XRAY-to-DRR translation enhances the registration results, by increasing the capture range and decreasing dependence on the similarity measure choice since the multi-modal registration becomes mono-modal.

A Review of Recent Advances in Deep Learning Models for Chest Disease Detection Using Radiography

Chest X-ray radiography (CXR) is among the most frequently used medical imaging modalities. It has a preeminent value in the detection of multiple life-threatening diseases. Radiologists can visually inspect CXR images for the presence of diseases. Most thoracic diseases have very similar patterns, which makes diagnosis prone to human error and leads to misdiagnosis. Computer-aided detection (CAD) of lung diseases in CXR images is among the popular topics in medical imaging research. Machine learning (ML) and deep learning (DL) provided techniques to make this task more efficient and faster. Numerous experiments in the diagnosis of various diseases proved the potential of these techniques. In comparison to previous reviews our study describes in detail several publicly available CXR datasets for different diseases. It presents an overview of recent deep learning models using CXR images to detect chest diseases such as VGG, ResNet, DenseNet, Inception, EfficientNet, RetinaNet, and ensemble learning methods that combine multiple models. It summarizes the techniques used for CXR image preprocessing (enhancement, segmentation, bone suppression, and data-augmentation) to improve image quality and address data imbalance issues, as well as the use of DL models to speed-up the diagnosis process. This review also discusses the challenges present in the published literature and highlights the importance of interpretability and explainability to better understand the DL models' detections. In addition, it outlines a direction for researchers to help develop more effective models for early and automatic detection of chest diseases.

AI-based radiodiagnosis using chest X-rays: A review

Chest Radiograph or Chest X-ray (CXR) is a common, fast, non-invasive, relatively cheap radiological examination method in medical sciences. CXRs can aid in diagnosing many lung ailments such as Pneumonia, Tuberculosis, Pneumoconiosis, COVID-19, and lung cancer. Apart from other radiological examinations, every year, 2 billion CXRs are performed worldwide. However, the availability of the workforce to handle this amount of workload in hospitals is cumbersome, particularly in developing and low-income nations. Recent advances in AI, particularly in computer vision, have drawn attention to solving challenging medical image analysis problems. Healthcare is one of the areas where AI/ML-based assistive screening/diagnostic aid can play a crucial part in social welfare. However, it faces multiple challenges, such as small sample space, data privacy, poor quality samples, adversarial attacks and most importantly, the model interpretability for reliability on machine intelligence. This paper provides a structured review of the CXR-based analysis for different tasks, lung diseases and, in particular, the challenges faced by AI/ML-based systems for diagnosis. Further, we provide an overview of existing datasets, evaluation metrics for different[][15mm][0mm]Q5 tasks and patents issued. We also present key challenges and open problems in this research domain.

Placenta segmentation in ultrasound imaging: Addressing sources of uncertainty and limited field-of-view

Automatic segmentation of the placenta in fetal ultrasound (US) is challenging due to the (i) high diversity of placenta appearance, (ii) the restricted quality in US resulting in highly variable reference annotations, and (iii) the limited field-of-view of US prohibiting whole placenta assessment at late gestation. In this work, we address these three challenges with a multi-task learning approach that combines the classification of placental location (e.g., anterior, posterior) and semantic placenta segmentation in a single convolutional neural network. Through the classification task the model can learn from larger and more diverse datasets while improving the accuracy of the segmentation task in particular in limited training set conditions. With this approach we investigate the variability in annotations from multiple raters and show that our automatic segmentations (Dice of 0.86 for anterior and 0.83 for posterior placentas) achieve human-level performance as compared to intra- and inter-observer variability. Lastly, our approach can deliver whole placenta segmentation using a multi-view US acquisition pipeline consisting of three stages: multi-probe image acquisition, image fusion and image segmentation. This results in high quality segmentation of larger structures such as the placenta in US with reduced image artifacts which are beyond the field-of-view of single probes.

Spatial feature and resolution maximization GAN for bone suppression in chest radiographs

BACKGROUND AND OBJECTIVE

Chest radiographs (CXR) are in great demand for visualizing the pathology of the lungs. However, the appearance of bones in the lung region hinders the localization of any lesion or nodule present in the CXR. Thus, bone suppression becomes an important task for the effective screening of lung diseases. Simultaneously, it is equally important to preserve spatial information and image quality because they provide crucial insights on the size and area of infection, color accuracy, structural quality, etc. Many researchers considered bone suppression as an image denoising problem and proposed conditional Generative Adversarial Network-based (cGAN) models for generating bone suppressed images from CXRs. These works do not focus on the retention of spatial features and image quality. The authors of this manuscript developed the Spatial Feature and Resolution Maximization (SFRM) GAN to efficiently minimize the visibility of bones in CXRs while ensuring maximum retention of critical information.

METHOD

This task is achieved by modifying the architectures of the discriminator and generator of the pix2pix model. The discriminator is combined with the Wasserstein GAN with Gradient Penalty to increase its performance and training stability. For the generator, a combination of different task-specific loss functions, viz., L1, Perceptual, and Sobel loss are employed to capture the intrinsic information in the image.

RESULT

The proposed model reported as measures of performance a mean PSNR of 43.588, mean NMSE of 0.00025, mean SSIM of 0.989, and mean Entropy of 0.454 bits/pixel on a test size of 100 images. Further, the combination of δ=10⁴, α=1, β=10, and γ=10 are the hyperparameters that provided the best trade-off between image denoising and quality retention.

CONCLUSION

The degree of bone suppression and spatial information preservation can be improved by adding the Sobel and Perceptual loss respectively. SFRM-GAN not only suppresses bones but also retains the image quality and intrinsic information. Based on the results of student's t-test it is concluded that SFRM-GAN yields statistically significant results at a 0.95 level of confidence and shows its supremacy over the state-of-the-art models. Thus, it may be used for denoising and preprocessing of images.

C2 -GAN: Content-consistent generative adversarial networks for unsupervised domain adaptation in medical image segmentation

PURPOSE

In clinical practice, medical image analysis has played a key role in disease diagnosis. One of the important steps is to perform an accurate organ or tissue segmentation for assisting medical professionals in making correct diagnoses. Despite the tremendous progress in the deep learning-based medical image segmentation approaches, they often fail to generalize to test datasets due to distribution discrepancies across domains. Recent advances aligning the domain gaps by using bi-directional GANs (e.g., CycleGAN) have shown promising results, but the strict constraints of the cycle consistency hamper these methods from yielding better performance. The purpose of this study is to propose a novel bi-directional GAN-based segmentation model with fewer constraints on the cycle consistency to improve the generalized segmentation results.

METHODS

We propose a novel unsupervised domain adaptation approach by designing content-consistent generative adversarial networks (C² -GAN) for medical image segmentation. Firstly, we introduce content consistency instead of cycle consistency to relax the constraint of the invertibility map to encourage the synthetic domain generated with a large domain transportation distance. The synthetic domain is thus pulled close to the target domain for the reduction of domain discrepancy. Secondly, we suggest a novel style transfer loss based on the difference in low-frequency magnitude to further mitigate the appearance shifts across domains.

RESULTS

We validate our proposed approach on three public X-ray datasets including the Montgomery, JSRT, and Shenzhen datasets. For an accurate evaluation, we randomly divided the images of each dataset into 70% for training, 10% for evaluation, and 20% for testing. The mean Dice was 95.73 ± 0.22%, 95.16 ± 1.42% for JSRT and Shenzhen datasets, respectively. For the recall and precision metrics, our model also achieved better or comparable performance than the state-of-the-art CycleGAN-based UDA approaches.

CONCLUSIONS

The experimental results validate the effectiveness of our method in mitigating the domain gaps and improving generalized segmentation results for X-ray image segmentation. This article is protected by copyright. All rights reserved.

Image enhancement techniques on deep learning approaches for automated diagnosis of COVID-19 features using CXR images

The outbreak of novel coronavirus (COVID-19) disease has infected more than 135.6 million people globally. For its early diagnosis, researchers consider chest X-ray examinations as a standard screening technique in addition to RT-PCR test. Majority of research work till date focused only on application of deep learning approaches that is relevant but lacking in better pre-processing of CXR images. Towards this direction, this study aims to explore cumulative effects of image denoising and enhancement approaches on the performance of deep learning approaches. Regarding pre-processing, suitable methods for X-ray images, Histogram equalization, CLAHE and gamma correction have been tested individually and along with adaptive median filter, median filter, total variation filter and gaussian denoising filters. Proposed study compared eleven combinations in exploration of most coherent approach in greedy manner. For more robust analysis, we compared ten CNN architectures for performance evaluation with and without enhancement approaches. These models are InceptionV3, InceptionResNetV2, MobileNet, MobileNetV2, Vgg19, NASNetMobile, ResNet101, DenseNet121, DenseNet169, DenseNet201. These models are trained in 4-way (COVID-19 pneumonia vs Viral vs Bacterial pneumonia vs Normal) and 3-way classification scenario (COVID-19 vs Pneumonia vs Normal) on two benchmark datasets. The proposed methodology determines with TVF + Gamma, models achieve higher classification accuracy and sensitivity. In 4-way classification MobileNet with TVF + Gamma achieves top accuracy of 93.25% with 1.91% improvement in accuracy score, COVID-19 sensitivity of 98.72% and F1-score of 92.14%. In 3-way classification our DenseNet201 with TVF + Gamma gains accuracy of 91.10% with improvement of 1.47%, COVID-19 sensitivity of 100% and F1-score of 91.09%. Proposed study concludes that deep learning modes with gamma correction and TVF + Gamma has superior performance compared to state-of-the-art models. This not only minimizes overlapping between COVID-19 and virus pneumonia but advantageous in time required to converge best possible results.

Enhanced Automatic Morphometry of Nerve Histological Sections Using Ensemble Learning

There is a need for an automated morphometry algorithm to facilitate the otherwise labor-intensive task of the quantitative histological analysis of neural microscopic images. A benchmark morphometry algorithm is the convolutional neural network Axondeepseg (ADS), which yields a high segmentation accuracy for scanning and transmission electron microscopy images. Nevertheless, it shows decreased accuracy when applied to optical microscopy images, and it has been observed to yield sizable false positives when identifying small-sized neurons within the slides. In this study, ensemble learning is used to enhance the performance of ADS by combining it with the paired image-to-image translation algorithm PairedImageTranslation (PIT). Here, 120 optical microscopy images of peripheral nerves were used to train and test the ensemble learning model and the two base models individually for comparison. The results showed weighted pixel-wise accuracy for the ensemble model of 95.5%, whereas the ADS and PIT yielded accuracies of 93.4% and 90%, respectively. The automated measurements of the axon diameters and myelin thicknesses from the manually marked ground truth images were not statistically different (p = 0.05) from the measurements taken from the same images when segmented using the developed ensemble model, while they were different when they were measured from the segmented images by the two base models individually. The automated measurement of the G ratios indicated a higher similarity to the ground truth testing images for the ensemble model in comparison with the individual base models. The proposed model yielded automated segmentation of the nerve slides, which were sufficiently equivalent to the manual annotations and could be employed for axon diameters and myelin thickness measurements for fully automated histological analysis of the neural images.

Generative adversarial networks and its applications in the biomedical image segmentation: a comprehensive survey

Recent advancements with deep generative models have proven significant potential in the task of image synthesis, detection, segmentation, and classification. Segmenting the medical images is considered a primary challenge in the biomedical imaging field. There have been various GANs-based models proposed in the literature to resolve medical segmentation challenges. Our research outcome has identified 151 papers; after the twofold screening, 138 papers are selected for the final survey. A comprehensive survey is conducted on GANs network application to medical image segmentation, primarily focused on various GANs-based models, performance metrics, loss function, datasets, augmentation methods, paper implementation, and source codes. Secondly, this paper provides a detailed overview of GANs network application in different human diseases segmentation. We conclude our research with critical discussion, limitations of GANs, and suggestions for future directions. We hope this survey is beneficial and increases awareness of GANs network implementations for biomedical image segmentation tasks.

U-shaped GAN for Semi-Supervised Learning and Unsupervised Domain Adaptation in High Resolution Chest Radiograph Segmentation

Deep learning has achieved considerable success in medical image segmentation. However, applying deep learning in clinical environments often involves two problems: (1) scarcity of annotated data as data annotation is time-consuming and (2) varying attributes of different datasets due to domain shift. To address these problems, we propose an improved generative adversarial network (GAN) segmentation model, called U-shaped GAN, for limited-annotated chest radiograph datasets. The semi-supervised learning approach and unsupervised domain adaptation (UDA) approach are modeled into a unified framework for effective segmentation. We improve GAN by replacing the traditional discriminator with a U-shaped net, which predicts each pixel a label. The proposed U-shaped net is designed with high resolution radiographs (1,024 × 1,024) for effective segmentation while taking computational burden into account. The pointwise convolution is applied to U-shaped GAN for dimensionality reduction, which decreases the number of feature maps while retaining their salient features. Moreover, we design the U-shaped net with a pretrained ResNet-50 as an encoder to reduce the computational burden of training the encoder from scratch. A semi-supervised learning approach is proposed learning from limited annotated data while exploiting additional unannotated data with a pixel-level loss. U-shaped GAN is extended to UDA by taking the source and target domain data as the annotated data and the unannotated data in the semi-supervised learning approach, respectively. Compared to the previous models dealing with the aforementioned problems separately, U-shaped GAN is compatible with varying data distributions of multiple medical centers, with efficient training and optimizing performance. U-shaped GAN can be generalized to chest radiograph segmentation for clinical deployment. We evaluate U-shaped GAN with two chest radiograph datasets. U-shaped GAN is shown to significantly outperform the state-of-the-art models.

Deep learning-based bone suppression in chest radiographs using CT-derived features: a feasibility study

Background

Bone suppression of chest X-ray holds the potential to improve the accuracy of target localization in image-guided radiation therapy (IGRT). However, the training dataset for bone suppression is limited because of the scarcity of bone-free radiographs. This study aims to develop a deep learning-based bone suppression method using CT-derived features to reduce the reliance on the bone-free dataset.

Methods

In this study, 59 high-resolution lung CT scans were processed to generate the lung digital radiographs (DRs), bone DRs, and bone-free DRs, for the training and internal validation of the proposed cascade convolutional neural network (CCNN). A three-stage image processing framework (CT segmentation, DR simulation, and feature expansion) was developed to expand simulated lung DRs with different weightings of bone intensity. The CCNN consists of a bone detection network and a bone suppression network. In external validation, the trained CCNN was evaluated using 30 chest radiographs. The synthesized bone-suppressed radiographs were compared with the bone-suppressed reference in terms of peak signal-to-noise ratio (PSNR), mean absolute error (MAE), structural similarity index measure (SSIM), and Spearman's correlation coefficient. Furthermore, the effectiveness of the proposed feature expansion method and CCNN model were assessed via the ablation experiment and replacement experiment, respectively.

Results

Evaluation on real chest radiographs showed that the bone-suppressed chest radiographs closely matched with the bone-suppressed reference, achieving an accuracy of MAE =0.0087±0.0030, SSIM =0.8458±0.0317, correlation of 0.9554±0.0170, and PNSR of 20.86±1.60. After removing the feature expansion from the CCNN model, the performance decreased in terms of MAE (0.0294±0.0093, -237.9%), SSIM (0.7747±0.0.0416, -8.4%), correlation (0.8772±0.0271, -8.2%), and PSNR (15.53±1.42, -25.5%) metrics.

Conclusions

We successfully demonstrated a novel deep learning-based bone suppression method using CT-derived features to reduce the reliance on the bone-free dataset. Implementation of the feature expansion procedures resulted in a remarkable reinforcement of the model performance. For the application of target localization in IGRT, the clinical testing of the proposed method in the context of radiation therapy is a necessary procedure to move from theory into practice.

Multi-label segmentation and detection of COVID-19 abnormalities from chest radiographs using deep learning

Due to COVID-19, demand for Chest Radiographs (CXRs) have increased exponentially. Therefore, we present a novel fully automatic modified Attention U-Net (CXAU-Net) multi-class segmentation deep model that can detect common findings of COVID-19 in CXR images. The architectural design of this model includes three novelties: first, an Attention U-net model with channel and spatial attention blocks is designed that precisely localize multiple pathologies; second, dilated convolution applied improves the sensitivity of the model to foreground pixels with additional receptive fields valuation, and third a newly proposed hybrid loss function combines both area and size information for optimizing model. The proposed model achieves average accuracy, DSC, and Jaccard index scores of 0.951, 0.993, 0.984, and 0.921, 0.985, 0.973 for image-based and patch-based approaches respectively for multi-class segmentation on Chest X-ray 14 dataset. Also, average DSC and Jaccard index scores of 0.998, 0.989 are achieved for binary-class segmentation on the Japanese Society of Radiological Technology (JSRT) CXR dataset. These results illustrate that the proposed model outperformed the state-of-the-art segmentation methods.

Automatic multi-plaque tracking and segmentation in ultrasonic videos

Carotid plaque tracking and segmentation in ultrasound videos is the premise for subsequent plaque property evaluation and treatment plan development. However, the task is quite challenging, as it needs to address the problems of poor image quality, plaque shape variations among frames, the existence of multiple plaques, etc. To overcome these challenges, we propose a new automatic multi-plaque tracking and segmentation (AMPTS) framework. AMPTS consists of three modules. The first module is a multi-object detector, in which a Dual Attention U-Net is proposed to detect multiple plaques and vessels simultaneously. The second module is a set of single-object trackers that can utilize the previous tracking results efficiently and achieve stable tracking of the current target by using channel attention and a ranking strategy. To make the first module and the second module work together, a parallel tracking module based on a simplified 'tracking-by-detection' mechanism is proposed to solve the challenge of tracking object variation. Extensive experiments are conducted to compare the proposed method with several state-of-the-art deep learning based methods. The experimental results demonstrate that the proposed method has high accuracy and generalizability with a Dice similarity coefficient of 0.83 which is 0.16, 0.06 and 0.27 greater than MAST (Lai et al., 2020), Track R-CNN (Voigtlaender et al., 2019) and VSD (Yang et al., 2019) respectively and has made significant improvements on seven other indicators. In the additional Testing set 2, our method achieved a Dice similarity coefficient of 0.80, an accuracy of 0.79, a precision of 0.91, a Recall 0.70, a F1 score of 0.79, an AP@0.5 of 0.92, an AP@0.7 of 0.74, and an expected average overlap of 0.79. Numerous ablation studies suggest the effectiveness of each proposed component and the great potential for multiple carotid plaques tracking and segmentation in clinical practice.

Accurate Estimation of Total Intracranial Volume in MRI using a Multi-tasked Image-to-Image Translation Network

Total intracranial volume (TIV) is the volume enclosed inside the cranium, inclusive of the meninges and the brain. TIV is extensively used to correct variations in inter-subject head size for the evaluation of neurodegenerative diseases. In this work, we present an automatic method to generate a TIV mask from MR images while synthesizing a CT image to be used in subsequent analysis. In addition, we propose an alternative way to obtain ground truth TIV masks using a semi-manual approach, which results in significant time savings. We train a conditional generative adversarial network (cGAN) using 2D MR slices to realize our tasks. The quantitative evaluation showed that the model was able to synthesize CT and generate TIV masks that closely approximate the reference images. This study also provides a comparison of the described method against skull stripping tools that output a mask enclosing the cranial volume, using MRI scan. In particular, highlighting the deficiencies in using such tools to approximate the volume using MRI scan.

Deep learning for chest X-ray analysis: A survey

Recent advances in deep learning have led to a promising performance in many medical image analysis tasks. As the most commonly performed radiological exam, chest radiographs are a particularly important modality for which a variety of applications have been researched. The release of multiple, large, publicly available chest X-ray datasets in recent years has encouraged research interest and boosted the number of publications. In this paper, we review all studies using deep learning on chest radiographs published before March 2021, categorizing works by task: image-level prediction (classification and regression), segmentation, localization, image generation and domain adaptation. Detailed descriptions of all publicly available datasets are included and commercial systems in the field are described. A comprehensive discussion of the current state of the art is provided, including caveats on the use of public datasets, the requirements of clinically useful systems and gaps in the current literature.

CheXLocNet: Automatic localization of pneumothorax in chest radiographs using deep convolutional neural networks

BACKGROUND

Pneumothorax can lead to a life-threatening emergency. The experienced radiologists can offer precise diagnosis according to the chest radiographs. The localization of the pneumothorax lesions will help to quickly diagnose, which will be benefit for the patients in the underdevelopment areas lack of the experienced radiologists. In recent years, with the development of large neural network architectures and medical imaging datasets, deep learning methods have become a methodology of choice for analyzing medical images. The objective of this study was to the construct convolutional neural networks to localize the pneumothorax lesions in chest radiographs.

METHODS AND FINDINGS

We developed a convolutional neural network, called CheXLocNet, for the segmentation of pneumothorax lesions. The SIIM-ACR Pneumothorax Segmentation dataset was used to train and validate CheXLocNets. The training dataset contained 2079 radiographs with the annotated lesion areas. We trained six CheXLocNets with various hyperparameters. Another 300 annotated radiographs were used to select parameters of these CheXLocNets as the validation set. We determined the optimal parameters by the AP50 (average precision at the intersection over union (IoU) equal to 0.50), a segmentation evaluation metric used by several well-known competitions. Then CheXLocNets were evaluated by a test set (1082 normal radiographs and 290 disease radiographs), based on the classification metrics: area under the receiver operating characteristic curve (AUC), sensitivity, specificity, and positive predictive value (PPV); segmentation metrics: IoU and Dice score. For the classification, CheXLocNet with best sensitivity produced an AUC of 0.87, sensitivity of 0.78 (95% CI 0.73-0.83), and specificity of 0.78 (95% CI 0.76-0.81). CheXLocNet with best specificity produced an AUC of 0.79, sensitivity of 0.46 (95% CI 0.40-0.52), and specificity of 0.92 (95% CI 0.90-0.94). For the segmentation, CheXLocNet with best sensitivity produced an IoU of 0.69 and Dice score of 0.72. CheXLocNet with best specificity produced an IoU of 0.77 and Dice score of 0.79. We combined them to form an ensemble CheXLocNet. The ensemble CheXLocNet produced an IoU of 0.81 and Dice score of 0.82. Our CheXLocNet succeeded in automatically detecting pneumothorax lesions, without any human guidance.

CONCLUSIONS

In this study, we proposed a deep learning network, called, CheXLocNet, for the automatic segmentation of chest radiographs to detect pneumothorax. Our CheXLocNets generated accurate classification results and high-quality segmentation masks for the pneumothorax at the same time. This technology has the potential to improve healthcare delivery and increase access to chest radiograph expertise for the detection of diseases. Furthermore, the segmentation results can offer comprehensive geometric information of lesions, which can benefit monitoring the sequential development of lesions with high accuracy. Thus, CheXLocNets can be further extended to be a reliable clinical decision support tool. Although we used transfer learning in training CheXLocNet, the parameters of CheXLocNet was still large for the radiograph dataset. Further work is necessary to prune CheXLocNet suitable for the radiograph dataset.

Dilated conditional GAN for bone suppression in chest radiographs with enforced semantic features

PURPOSE

The purpose of this essay is to improve computer-aided diagnosis of lung diseases by the removal of bone structures imagery such as ribs and clavicles, which may shadow a clinical view of lesions. This paper aims to develop an algorithm to suppress the imaging of bone structures within clinical x-ray images, leaving a residual portrayal of lung tissue; such that these images can be used to better serve applications, such as lung nodule detection or pathology based on the radiological reading of chest x rays.

METHODS

We propose a conditional Adversarial Generative Network (cGAN) (Mirza and Osindero, Conditional generative adversarial nets, 2014.) model, consisting of a generator and a discriminator, for the task of bone shadow suppression. The proposed model utilizes convolutional operations to expand the size of the receptive field of the generator without losing contextual information while downsampling the image. It is trained by enforcing both the pixel-wise intensity similarity and the semantic-level visual similarity between the generated x-ray images and the ground truth, via optimizing an L-1 loss of the pixel intensity values on the generator side and a feature matching loss on the discriminator side, respectively.

RESULTS

The framework we propose is trained and tested on an open-access chest radiograph dataset for benchmark. Results show that our model is capable of generating bone-suppressed images of outstanding quality with a limited number of training samples (N = 272).

CONCLUSIONS

Our approach outperforms current state-of-the-art bone suppression methods using x-ray images. Instead of simply downsampling images at different scales, our proposed method mitigates the loss of contextual information by utilizing dilated convolutions, which gains a noticeable quality improvement for the outputs. On the other hand, our experiment shows that enforcing the semantic similarity between the generated and the ground truth images assists the adversarial training process and achieves better perceptual quality.