Multi-task contrastive learning for automatic CT and X-ray diagnosis of COVID-19

Thin-section multiplanar reformats from multidetector-row CT data: Utility for assessment of regional tumor extent in non-small cell lung cancer

Significant work has been done towards deep learning (DL) models for automatic lung and lesion segmentation and classification of COVID-19 on chest CT data. However, comprehensive visualization systems focused on supporting the dual visual+DL diagnosis of COVID-19 are non-existent. We present COVID-view, a visualization application specially tailored for radiologists to diagnose COVID-19 from chest CT data. The system incorporates a complete pipeline of automatic lungs segmentation, localization/isolation of lung abnormalities, followed by visualization, visual and DL analysis, and measurement/quantification tools. Our system combines the traditional 2D workflow of radiologists with newer 2D and 3D visualization techniques with DL support for a more comprehensive diagnosis. COVID-view incorporates a novel DL model for classifying the patients into positive/negative COVID-19 cases, which acts as a reading aid for the radiologist using COVID-view and provides the attention heatmap as an explainable DL for the model output. We designed and evaluated COVID-view through suggestions, close feedback and conducting case studies of real-world patient data by expert radiologists who have substantial experience diagnosing chest CT scans for COVID-19, pulmonary embolism, and other forms of lung infections. We present requirements and task analysis for the diagnosis of COVID-19 that motivate our design choices and results in a practical system which is capable of handling real-world patient cases.

Coronavirus disease (COVID-19) is highly contagious and pathogenic. Currently, the diagnosis of COVID-19 is based on nucleic acid testing, but it has false negatives and hysteresis. The use of lung CT scans can help screen and effectively monitor diagnosed cases. The application of computer-aided diagnosis technology can reduce the burden on doctors, which is conducive to rapid and large-scale diagnostic screening. In this paper, we proposed an automatic detection method for COVID-19 based on spatiotemporal information fusion. Using the segmentation network in the deep learning method to segment the lung area and the lesion area, the spatiotemporal information features of multiple CT scans are extracted to perform auxiliary diagnosis analysis. The performance of this method was verified on the collected dataset. We achieved the classification of COVID-19 CT scans and non-COVID-19 CT scans and analyzed the development of the patients' condition through the CT scans. The average accuracy rate is 96.7%, sensitivity is 95.2%, and F1 score is 95.9%. Each scan takes about 30 seconds for detection.

In computed tomography (CT) cardiovascular imaging, the numerous contrast injection protocols used to enhance structures make it difficult to gather training datasets for deep learning applications supporting diverse protocols. Moreover, creating annotations on noncontrast scans is extremely tedious. Recently, spectral CT's virtual-noncontrast images (VNC) have been used as data augmentation to train segmentation networks performing on enhanced and true-noncontrast (TNC) scans alike, while improving results on protocols absent of their training dataset. However, spectral data are not widely available, making it difficult to gather specific datasets for each task. As a solution, we present a data augmentation workflow based on a trained image translation network, to bring spectral-like augmentation to any conventional CT dataset. The conventional CT-to-spectral image translation network (HUSpectNet) was first trained to generate VNC from conventional housnfied units images (HU), using an unannotated spectral dataset of 1830 patients. It was then tested on a second dataset of 300 spectral CT scans by comparing VNC generated through deep learning (VNCDL ) to their true counterparts. To illustrate and compare our workflow's efficiency with true spectral augmentation, HUSpectNet was applied to a third dataset of 112 spectral scans to generate VNCDL along HU and VNC images. Three different three-dimensional (3D) networks (U-Net, X-Net, and U-Net++) were trained for multilabel heart segmentation, following four augmentation strategies. As baselines, trainings were performed on contrasted images without (HUonly) and with conventional gray-values augmentation (HUaug). Then, the same networks were trained using a proportion of contrasted and VNC/VNCDL images (TrueSpec/GenSpec). Each training strategy applied to each architecture was evaluated using Dice coefficients on a fourth multicentric multivendor single-energy CT dataset of 121 patients, including different contrast injection protocols and unenhanced scans. The U-Net++ results were further explored with distance metrics on every label. Tested on 300 full scans, our HUSpectNet translation network shows a mean absolute error of 6.70±2.83 HU between VNCDL and VNC, while peak signal-to-noise ratio reaches 43.89dB. GenSpec and TrueSpec show very close results regardless of the protocol and used architecture: mean Dice coefficients (DSCmean ) are equal with a margin of 0.006, ranging from 0.879 to 0.938. Their performances significantly increase on TNC scans (p-values<0.017 for all architectures) compared to HUonly and HUaug, with DSCmean of 0.448/0.770/0.879/0.885 for HUonly/HUaug/TrueSpec/GenSpec using the U-Net++ architecture. Significant improvements are also noted for all architectures on chest-abdominal-pelvic scans (p-values<0.007) compared to HUonly and for pulmonary embolism scans (p-values<0.039) compared to HUaug. Using U-Net++, DSCmean reaches 0.892/0.901/0.903 for HUonly/TrueSpec/GenSpec on pulmonary embolism scans and 0.872/0.896/0.896 for HUonly/TrueSpec/GenSpec on chest-abdominal-pelvic scans. Using the proposed workflow, we trained versatile heart segmentation networks on a dataset of conventional enhanced CT scans, providing robust predictions on both enhanced scans with different contrast injection protocols and TNC scans. The performances obtained were not significantly inferior to training the model on a genuine spectral CT dataset, regardless of the architecture implemented. Using a general-purpose conventional-to-spectral CT translation network as data augmentation could therefore contribute to reducing data collection and annotation requirements for machine learning-based CT studies, while extending their range of application.

The main aim of this work was to perform rigid registration of Computed Tomography (CT) and scanner datasets. The surgeon applies CT and scanner datasets in computer aided surgery and performs registration in order to visualize the location of surgical instrument on screen. It is well known fact that the registration procedure is crucial for efficient computer aiding of surgery. Selected algorithm should take into account types of datasets, required accuracy and time of calculations. The algorithms are classified basing on the various criteria: e.g. precision (coarse and fine registration), types of pointset (set of pair of corresponding points – so called point-point method, unorganized sets of points – so called surface registration). The paper presents exemplary results of applying the following algorithms: Landmark Transform (point-point registration), two methods of uninitialized Iterative Closest Point type (surface registration) and a hybrid method. The evaluated factors were: distance error (mean, minimal and maximal value) and running time of algorithm. The algorithms were tested on various datasets: (1) two similar datasets from Computed Tomography (one is geometrically transformed), (2) Computed Tomography dataset and cloud of points recorded using 3D Artec Space Spider scanner. In the first case the mean error values equaled: 102.08 mm – 121.70 mm for uninitialized ICPs methods, 0.005 mm for Landmark Transform method, and 0.0003 mm for hybrid method. The slowest algorithms in our tests were ICPs methods, faster was hybrid algorithm, and the fastest was Landmark Transform method. In the second case the distance errors were evaluated in four selected points, and the smallest errors were: 23.21 mm for uninitialized ICPs method, 0.69 mm for Landmark Transform, 9.03 for hybrid method. All algorithms were relatively slow for these large datasets, the fastest was Landmark Transform. In the second part of research we analysed the Target Registration Error (TRE) for fused Computed Tomography and scanner-recorded dataset. The TRE values equaled 0.7 mm - 2.8 mm. The results of CT – scanner datasets registration highly depend on the similarity of sets, especially their overlapping, but also their resolutions and uniformities.

AI-based wavelet and stacked deep learning architecture for detecting coronavirus (COVID-19) from chest X-ray images

COVID-DSNet: A novel deep convolutional neural network for detection of coronavirus (SARS-CoV-2) cases from CT and Chest X-Ray images

Objective: To demonstrate the possibilities, advantages and limitations of virtual bronchoscopy using data sets from positron emission tomography (PET) and computed tomography (CT).Materials and Methods: Eight consecutive patients with non-small cell lung cancer (NSCLC) underwent PET/CT. PET was performed with a glucose analog, 2-[fluorine-18]-fluoro-2-deoxy-D-glucose (18F-FDG), using a state-of-the-art full-ring Pico-3D PET scanner. CT was performed with a venous-dominant contrast-enhanced phase using a 16-slice CT scanner. The tracheobronchial system was segmented using the CT data set with an interactive threshold interval volume-growing segmentation algorithm. The primary tumors and lymph node metastases were segmented for virtual CT-bronchoscopy using the CT data set and for virtual hybrid bronchoscopy using the PET/CT data set. The structures of interest were visualized with a color-coded shaded-surface rendering method.Results: The use of CT and virtual CT-bronchoscopy primarily facilitates visualization of the anatomical details of the tracheobronchial system and detection of anatomical/morphologic structural changes caused by disease. PET/CT and virtual hybrid bronchoscopy, or virtual PET/CT-bronchoscopy, give superior results to virtual CT-bronchoscopy because the hybrid bronchoscopy uses both the CT information and the molecular/metabolic information about the disease obtained from PET.Conclusions: PET/CT imaging has proven to be a highly valuable oncological diagnostic modality. Virtual hybrid bronchoscopy can be performed using a low-dose CT scan or diagnostic CT. However, it is expected to improve diagnostic accuracy in identification and characterization of malignancies, verification of infections, and differentiation of viable tumor tissue from atelectases and scar tissue, as well as assessment of tumor staging and therapeutic response, and detection of early stage recurrences that are not detectable or are liable to be misjudged using virtual CT-bronchoscopy. It could also be useful as a screening examination method for patients with suspected endobronchial malignancy. Virtual hybrid bronchoscopy with a transparent color-coded shaded-surface rendering model offers a useful alternative to fiberoptic bronchoscopy, and is particularly promising for patients for whom fiberoptic bronchoscopy is not feasible, contraindicated or refused.

COVID-19 is a newly identified disease, which is very contagious and has been rapidly spreading across different countries around the world, calling for rapid and accurate diagnosis tools. Chest CT imaging has been widely used in clinical practice for disease diagnosis, but image reading is still a time-consuming work. We aim to integrate an image preprocessing technology for anomaly detection with supervised deep learning for chest CT imaging-based COVID-19 diagnosis. In this study, a matrix profile technique was introduced to CT image anomaly detection in two levels. At one-dimensional level, CT images were simply flatted and transformed to a one-dimensional vector so that the matrix profile algorithm could be implemented for them directly. At two-dimensional level,a matrix profile was calculated in a sliding window way for every segment in the image. An anomaly severity score (CT-SS) was calculated, and the difference of the CT-SS between the COVID-19 CT images and Non-COVID-19 CT images was tested. A sparse anomaly mask was calculated and applied to penalize the pixel values of each image. The anomaly weighted images were then used to train standard DenseNet deep learning models to distinguish the COVID-19 CT from Non-COVID-19 CT images. A VGG19 model was used as a baseline model for comparison. Although extra finetuning needs to be done manually, the one-dimensional matrix profile method could identify the anomalies successfully. Using the two-dimensional matrix profiling method, CT-SS and anomaly weighted image can be successfully generated for each image. The CT-SS significantly differed among the COVID-19 CT images and Non-COVID-19 CT images ($p-value <; 0.05$ ). Furthermore, we identified a potential causal association between the number of underlying diseases of a COVID-19 patient and the severity of the disease through statistical mediation analysis. Compared to the raw images, the anomaly weighted images showed generally better performance in training the DenseNet models with different architectures for diagnosing COVID-19, which was validated using two publicly available COVID-19 lung CT image datasets. The metric Area Under the Curve(AUC) on one dataset were 0.7799(weighted)vs. 0.7391(unweighted), 0.7812(weighted) vs. 0.7410(unweighted), 0.7780(weighted) vs. 0.7399(unweighted), 0.7045(weighted) vs. 0.6910(unweighted) for DenseNet121, DenseNet169, DenseNet201, and the baseline model VGG19, respectively. The same trend was observed using another independent dataset. The significant results revealed the critical value of using this existing state-of-the-art algorithm for image anomaly detection. Furthermore, the end-to-end model structure has the potential to work as a rapid tool for clinical imaging-based diagnosis.

To retrospectively determine whether alignment of registered positron emission tomographic (PET) and computed tomographic (CT) data sets obtained independently varies significantly from alignment of data sets acquired from a combined PET/CT scanner. The study was approved by the institution's Human Research Committee with a waiver of informed consent and complied with HIPAA. Whole-body combined PET/CT data sets and separate routinely positioned thoracic CT data sets were obtained from 12 patients (six men, six women; mean age, 48.6 years; range, 24-62 years). Separate PET and thoracic CT data sets matched for patient positioning and respiration were acquired on the same day for nine patients (four men, five women; mean age, 71 years; range, 51-90 years). Computer nonlinear registration was performed on PET and CT data sets from combined PET/CT (fusion group 1), PET data sets from combined PET/CT with unmatched thoracic CT (fusion group 2), and data sets from separate PET and CT matched for patient positioning and respiration (fusion group 3). Quality of alignment was assessed by two radiologists in consensus blinded to the source of registered data in each fusion group at the following anatomic locations: diaphragm, aortic arch, heart, thoracic spine, and lung apices. Results were compared by using the Wilcoxon paired signed rank and unpaired rank sum tests. Quality of alignment did not significantly differ between fusion group 1 and fusion group 3. Fusion group 1 provided significantly better alignment in two of five anatomic locations (P = .008 for diaphragm and P = .031 for heart) than fusion group 2. Fusion group 3 provided significantly better alignment in two of five anatomic locations (P = .037 for diaphragm and P = .009 for heart) than fusion group 2. Thoracic anatomic alignment does not significantly differ between registered PET and CT data sets acquired on a combined PET/CT scanner or from separate PET and CT scanners obtained on the same day when carefully matched for anatomic positioning and respiration.

More than 400 million cases of the new coronavirus (COVID-19) have been confirmed since December 2019 in more than 200 countries. Since the spread of original COVID-19 virus SARS-CoV-2, thousands of mutations have been discovered. The most dominant ones are Alpha, Beta, Gama, Delta and Omicron variants, with the Omicron variant rapidly spreading and dominating the current phase of the COVID wave across the globe. It needs early detection and self-isolation to contain the virus. Molecular tests like rRTPCR are common for its detection. However, with the current spreading rate and lack of availability of large-scale testing laboratories, rapid diagnosis has become difficult. COVID-19 diagnosis from CT and X-ray images using deep learning techniques has been the subject of a lot of research in the last two years. This work presents a review of these studies sourced from top databases such as Web of Science and highlights challenges and research gaps with future research directions.

Preoperative COVID-19 CT screening in renal transplant recipients

Liver segmentation from computed tomography (CT) images has grown significantly in importance in the field of medical image processing in the last few years. It is the first and most crucial step in any computerized technique for the automatic detection of liver disease, liver volume measurement, and 3D liver volume rendering. The diagnosis and treatment of liver cancer depend heavily on the segmentation of the liver from CT images to get liver volumetric data, but manual segmentation is a strenuous and time-consuming process. The procedure can be accelerated, simplified, and made less error-prone by using deep learning methods. Image segmentation based on deep learning techniques has gained widespread acceptance due to its robustness, efficiency, and it's reproducible nature. Therefore, in this paper, using UNet, MONAI (Medical Open Network for Artificial Intelligence) and PyTorch framework, a deep-learning model to segment the liver from publicly available CT scan dataset was developed. The same ideas that underlie this model for segmenting the liver will allow to create models for segmenting other organs or malignancies using CT data. The goal is to develop a liver segmentation model that can quickly and accurately extract the liver from any given CT image with an accuracy that is on par of manual segmentation performed by a skilled radiologist.

Coronavirus disease (COVID-19) constitutes an ongoing global health problem with significant morbidity and mortality. It usually presents characteristic findings on a chest CT scan, which may lead to early detection of the disease. A timely and accurate diagnosis of COVID-19 is the cornerstone for the prompt management of the patients. The aim of the present study was to evaluate the performance of an automated machine learning algorithm in the diagnosis of Covid-19 pneumonia using chest CT scans. Diagnostic performance was assessed by the area under the receiver operating characteristic curve (AUC), sensitivity, and positive predictive value. The method's average precision was 0.932. We suggest that auto-ML platforms help users with limited ML expertise train image recognition models by only uploading the examined dataset and performing some basic settings. Such methods could deliver significant potential benefits for patients in the future by allowing for earlier disease detection and care.

Classification of lung nodule malignancy in computed tomography imaging utilising generative adversarial networks and semi-supervised transfer learning