Original CT Images Research Articles

Lumbar spine localisation method based on feature fusion

AbstractTo eliminate unnecessary background information, such as soft tissues in original CT images and the adverse impact of the similarity of adjacent spines on lumbar image segmentation and surgical path planning, a two‐stage approach for localising lumbar segments is proposed. First, based on the multi‐scale feature fusion technology, a non‐linear regression method is used to achieve accurate localisation of the overall spatial region of the lumbar spine, effectively eliminating useless background information, such as soft tissues. In the second stage, we directly realised the precise positioning of each segment in the lumbar spine space region based on the non‐linear regression method, thus effectively eliminating the interference caused by the adjacent spine. The 3D Intersection over Union (3D_IOU) is used as the main evaluation indicator for the positioning accuracy. On an open dataset, 3D_IOU values of 0.8339 ± 0.0990 and 0.8559 ± 0.0332 in the first and second stages, respectively is achieved. In addition, the average time required for the proposed method in the two stages is 0.3274 and 0.2105 s respectively. Therefore, the proposed method performs very well in terms of both precision and speed and can effectively improve the accuracy of lumbar image segmentation and the effect of surgical path planning.

Self-supervised region-aware segmentation of COVID-19 CT images using 3D GAN and contrastive learning

Medical image segmentation is a key initial step in several therapeutic applications. While most of the automatic segmentation models are supervised, which require a well-annotated paired dataset, we introduce a novel annotation-free pipeline to perform segmentation of COVID-19 CT images. Our pipeline consists of three main subtasks: automatically generating a 3D pseudo-mask in self-supervised mode using a generative adversarial network (GAN), leveraging the quality of the pseudo-mask, and building a multi-objective segmentation model to predict lesions. Our proposed 3D GAN architecture removes infected regions from COVID-19 images and generates synthesized healthy images while keeping the 3D structure of the lung the same. Then, a 3D pseudo-mask is generated by subtracting the synthesized healthy images from the original COVID-19 CT images. We enhanced pseudo-masks using a contrastive learning approach to build a region-aware segmentation model to focus more on the infected area. The final segmentation model can be used to predict lesions in COVID-19 CT images without any manual annotation at the pixel level. We show that our approach outperforms the existing state-of-the-art unsupervised and weakly-supervised segmentation techniques on three datasets by a reasonable margin. Specifically, our method improves the segmentation results for the CT images with low infection by increasing sensitivity by 20% and the dice score up to 4%. The proposed pipeline overcomes some of the major limitations of existing unsupervised segmentation approaches and opens up a novel horizon for different applications of medical image segmentation.

Detection and Classification of Bronchiectasis Based on Improved Mask-RCNN.

Bronchiectasis is defined as a permanent dilation of the bronchi that can cause pulmonary ventilation dysfunction. CT examination is an important means of diagnosing bronchiectasis. It can also be used in severity scoring. Current studies on bronchiectasis have focused on high-resolution CT (HRCT), ignoring the more common low-dose CT (LDCT). Methodologically, existing studies have not adopted an authoritative standard to classify the severity of bronchiectasis. In effect, the accuracy of detection and classification needs to be improved for practical application. In this paper, the ACER image enhancement method, RDU-Net lung lobe segmentation method and HDC Mask R-CNN model were proposed to detect and classify bronchiectasis. Moreover, a Python-based system was developed: after inputing an LDCT image of a patient’s lung, it can automatically perform a series of processing, then call on the trained deep learning model for detection and classification, and automatically obtain the patient’s bronchiectasis final score according to the Reiff and BRICS scoring criteria. In this paper, the mapping relationship between original lung CT image data and bronchiectasis scoring system was established. The accuracy of the method proposed in this paper was 91.4%; the IOU, sensitivity and specificity were 88.8%, 88.6% and 85.4%, respectively; and the recognition speed of one picture was about 1 s. Compared to a human doctor, the system can process large amounts of data simultaneously, quickly and efficiently, with the same judgment accuracy as a human doctor. Doctors only need to judge the uncertain cases, which significantly reduces the burden of doctors and provides a useful reference for doctors to diagnose the disease.

CT-Only Radiotherapy: An Exploratory Study for Automatic Dose Prediction on Rectal Cancer Patients Via Deep Adversarial Network.

PurposeCurrent deep learning methods for dose prediction require manual delineations of planning target volume (PTV) and organs at risk (OARs) besides the original CT images. Perceiving the time cost of manual contour delineation, we expect to explore the feasibility of accelerating the radiotherapy planning by leveraging only the CT images to produce high-quality dose distribution maps while generating the contour information automatically.Materials and MethodsWe developed a generative adversarial network (GAN) with multi-task learning (MTL) strategy to produce accurate dose distribution maps without manually delineated contours. To balance the relative importance of each task (i.e., the primary dose prediction task and the auxiliary tumor segmentation task), a multi-task loss function was employed. Our model was trained, validated and evaluated on a cohort of 130 rectal cancer patients.ResultsExperimental results manifest the feasibility and improvements of our contour-free method. Compared to other mainstream methods (i.e., U-net, DeepLabV3+, DoseNet, and GAN), the proposed method produces the leading performance with statistically significant improvements by achieving the highest HI of 1.023 (3.27E-5) and the lowest prediction error with ΔD95 of 0.125 (0.035) and ΔDmean of 0.023 (4.19E-4), respectively. The DVH differences between the predicted dose and the ideal dose are subtle and the errors in the difference maps are minimal. In addition, we conducted the ablation study to validate the effectiveness of each module. Furthermore, the results of attention maps also prove that our CT-only prediction model is capable of paying attention to both the target tumor (i.e., high dose distribution area) and the surrounding healthy tissues (i.e., low dose distribution areas).ConclusionThe proposed CT-only dose prediction framework is capable of producing acceptable dose maps and reducing the time and labor for manual delineation, thus having great clinical potential in providing accurate and accelerated radiotherapy. Code is available at https://github.com/joegit-code/DoseWithCT

Studying the relation between post-infarction ventricular arrhythmia and left ventricular myocardium thinning on computed tomography images using explainable deep learning

Abstract Funding Acknowledgements Type of funding sources: Public Institution(s). Main funding source(s): National Research Agency, 3IA Côte d’Azur (ANR-19-P3IA-0002), IHU Liryc (ANR- 10-IAHU-04) Background Infarct heterogeneity plays a critical role in the development of scar-related ventricular arrhythmias (VA). Wall thickness (WT) from CT was shown to correlate with arrhythmogenic sites in the context of ablation. Purpose To analyze the relationship between WT distribution and the presence VAs in patients with history of myocardial infarction (MI). Methods From 2010 to 2020, we retrospectively included consecutive patients who underwent CT more than 1 year after MI.Automated LV wall segmentation, reorientation, WT computation and flattening methods were applied to obtain 2D WT bullseye maps.The population was divided into a training set (3/4) and a testing set (1/4). On the training set, a conditional variational autoencoder (CVAE) model was trained to encode the WT map in its latent representation,which was then used by a classifier model to predict VA. For each prediction, a gradient back-projection method was used to generate attention maps highlighting the bullseye region most influential in the model’s decision. The ability of the trained CVAE to identify patients with VA was then assessed on the test population, and compared to that of other clinical variables (age, gender, LVEF, scar size). Results 641 patients were included (age 73±7 years, 83% males, LVEF 46±10%), including 166 (26%) with history of VA. From original CT images, automated processing methods allowed for the obtention of a WT bullseye, a VA prediction and an attention map in less than 2 min. On the testing population, univariable correlates of VA were LVEF (P<0.001), scar size defined as WT area (P<0.001), CVAE prediction (P<0.001), and male gender (P=0.007). Multivariable analysis identified CVAE prediction and male gender as independent VA correlates (P<0.001 and P=0.01, R2=0.364), while LVEF and scar size were not (P=0.052 and P=0.60). The CVAE model identified patients with VA with a sensitivity/specificity of 0.87/0.74. The analysis of attention maps confirmed that the CVAE actually used the MI region to predict VA, and showed that the CVAE was filtering out areas of physiological WT, thereby outperforming simple WT thresholding. Conclusions Fully automated analysis of WT from CT images is feasible in patients with MI, and allows for the extraction of robust markers of VA, outperforming conventional risk stratifiers

Streak Metal Artifact Reduction Based on Sinogram Fusion and Tissue-Class Model in CT Images

The presence of streak metal artifacts seriously degrades the diagnostic value and deteriorates the qualities of CT images. Analyzing the causes and classical streak metal artifact reduction (MAR) methods, the paper proposes the streak metal artifact reduction method based on sinogram fusion and tissue-class mode for CT images (F-MAR). Firstly, the original CT images are corrected using a linear interpolation streak metal artifact reduction (L-MAR) scheme in the raw data domain. Subsequently, to preserve the edge information, the metal artifact-reduced images are then smoothed into smoothed images (tissue-class model) by using the mean filter. Segment the original CT image that contained the streak artifacts. The original CT image and the CT image that contained high-density material are projected into the original sinogram and the high density material sinogram, respectively. Secondly, the simple linear interpolation is used to correct the CT original CT image into the corrected CT image. The mean filter is applied in the corrected CT image. The corrected CT image is projected into the corrected sinogram. Thirdly, according to the position of the high density material sinogram located in the original sinogram and the corrected sinogram, the original position sinogram included in the original sinogram and the corrected position sinogram included in the corrected sinogram are, respectively, obtained. The two sinograms are fused into the fused sinogram. The fused sinogram, the original sinogram, and the high-density material sinogram are fused into the final sinogram. Finally, the filtered back projection reconstruction algorithm is used to reconstruct the final sinogram into the reconstructed CT image. The reconstructed CT image and high density material image are fused into the final image. The experimentation results show that the method proposed in the paper can obtain better correction effect than the classical correction methods in vision.

Two nomograms for differentiating mass-forming chronic pancreatitis from pancreatic ductal adenocarcinoma in patients with chronic pancreatitis.

To develop and validate a CT nomogram and a radiomics nomogram to differentiate mass-forming chronic pancreatitis (MFCP) from pancreatic ductal adenocarcinoma (PDAC) in patients with chronic pancreatitis (CP). In this retrospective study, the data of 138 patients with histopathologically diagnosed MFCP or PDAC treated at our institution were retrospectively analyzed. Two radiologists analyzed the original cross-sectional CT images based on predefined criteria. Image segmentation, feature extraction, and feature reduction and selection were used to create the radiomics model. The CT and radiomics models were developed using data from a training cohort of 103 consecutive patients. The models were validated in 35 consecutive patients. Multivariable logistic regression analysis was conducted to develop a model for the differential diagnosis of MFCP and PDAC and visualized as a nomogram. The nomograms' performances were determined based on their differentiating ability and clinical utility. The mean age of patients was 53.7 years, 75.4% were male. The CT nomogram showed good differentiation between the two entities in the training (area under the curve [AUC], 0.87) and validation (AUC, 0.94) cohorts. The radiomics nomogram showed good differentiation in the training (AUC, 0.91) and validation (AUC, 0.93) cohorts. Decision curve analysis showed that patients could benefit from the CT and radiomics nomograms, if the threshold probability was 0.05-0.85 and > 0.05, respectively. The two nomograms reasonably accurately differentiated MFCP from PDAC in patients with CP and hold potential for refining the management of pancreatic masses in CP patients. • A CT nomogram and a computed tomography-based radiomics nomogram reasonably accurately differentiated mass-forming chronic pancreatitis from pancreatic ductal adenocarcinoma in patients with chronic pancreatitis (CP). • The two nomograms can monitor the cancer risk in patients with CP and hold promise to optimize the management of pancreatic masses in patients with CP.

Detection of Covid-19 on Localized Ct-scan Images Using Deep Learning Convolution Neural Network

Pneumonia Coronavirus Disease 2019 (COVID-19) is an inflammation of the lung parenchyma caused by Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2). Supporting examinations carried out to establish a diagnosis of Covid-19 is through radiological examinations, one of which is Computed Tomography Scan (CT-Scan). The current method used to diagnose COVID-19 from CT scan images is by studying the 2-D CT Scan image data set using naked eye, then interpreting the data one by one. This procedure is ineffective. Research aim is to develop Covid-19 detection application on localized CT-Scan images using Deep Learning Convolution Neural Network (CNN). Pre-Trained Model used is ResNet-50. Test was carried out 3 times. First test uses original CT image, with an accuracy of 92.5%. Second test used pulmonary CT data that had been separated from surrounding tissue, with an accuracy of 95%. The last test used localized CT data (Covid Candidate), the accuracy obtained was 98%.

Retrospective study of deep learning to reduce noise in non-contrast head CT images.

Presented herein is a novel CT denoising method uses a skip residual encoder-decoder framework with group convolutions and a novel loss function to improve the subjective and objective image quality for improved disease detection in patients with acute ischemic stroke (AIS). In this retrospective study, confirmed AIS patients with full-dose NCCT head scans were randomly selected from a stroke registry between 2016 and 2020. 325 patients (67±15 years, 176 men) were included. 18 patients each with 4-7 NCCTs performed within 5-day timeframe (83 total scans) were used for model training; 307 patients each with 1-4 NCCTs performed within 5-day timeframe (380 total scans) were used for hold-out testing. In the training group, a mean CT was created from the patient's co-registered scans for each input CT to train a rotation-reflection equivariant U-Net with skip and residual connections, as well as a group convolutional neural network (SRED-GCNN) using a custom loss function to remove image noise. Denoising performance was compared to the standard Block-matching and 3D filtering (BM3D) method and RED-CNN quantitatively and visually. Signal-to-noise ratio (SNR) and contrast-to-noise (CNR) were measured in manually drawn regions-of-interest in grey matter (GM), white matter (WM) and deep grey matter (DG). Visual comparison and impact on spatial resolution were assessed through phantom images. SRED-GCNN reduced the original CT image noise significantly better than BM3D, with SNR improvements in GM, WM, and DG by 2.47x, 2.83x, and 2.64x respectively and CNR improvements in DG/WM and GM/WM by 2.30x and 2.16x respectively. Compared to the proposed SRED-GCNN, RED-CNN reduces noise effectively though the results are visibly blurred. Scans denoised by the SRED-GCNN are shown to be visually clearer with preserved anatomy. The proposed SRED-GCNN model significantly reduces image noise and improves signal-to-noise and contrast-to-noise ratios in 380 unseen head NCCT cases.

Decoding intra-tumoral spatial heterogeneity on radiological images using the Hilbert curve

BackgroundCurrent intra-tumoral heterogeneous feature extraction in radiology is limited to the use of a single slice or the region of interest within a few context-associated slices, and the decoding of intra-tumoral spatial heterogeneity using whole tumor samples is rare. We aim to propose a mathematical model of space-filling curve-based spatial correspondence mapping to interpret intra-tumoral spatial locality and heterogeneity.MethodsA Hilbert curve-based approach was employed to decode and visualize intra-tumoral spatial heterogeneity by expanding the tumor volume to a two-dimensional (2D) matrix in voxels while preserving the spatial locality of the neighboring voxels. The proposed method was validated using three-dimensional (3D) volumes constructed from lung nodules from the LIDC-IDRI dataset, regular axial plane images, and 3D blocks.ResultsDimensionality reduction of the Hilbert volume with a single regular axial plane image showed a sparse and scattered pixel distribution on the corresponding 2D matrix. However, for 3D blocks and lung tumor inside the volume, the dimensionality reduction to the 2D matrix indicated regular and concentrated squares and rectangles. For classification into benign and malignant masses using lung nodules from the LIDC-IDRI dataset, the Inception-V4 indicated that the Hilbert matrix images improved accuracy (85.54% vs. 73.22%, p < 0.001) compared to the original CT images of the test dataset.ConclusionsOur study indicates that Hilbert curve-based spatial correspondence mapping is promising for decoding intra-tumoral spatial heterogeneity of partial or whole tumor samples on radiological images. This spatial-locality-preserving approach for voxel expansion enables existing radiomics and convolution neural networks to filter structured and spatially correlated high-dimensional intra-tumoral heterogeneity.

Damage characteristics of coal under different loading modes based on CT three-dimensional reconstruction

Due to the heterogeneous characteristics of coal materials, it is difficult to carry out repetitive tests and quantitative analysis of internal damage of coal. Therefore, the damage characteristics of coal under uniaxial and true triaxial loading was studied by numerical simulation based on 3D reconstruction modeling, and the corresponding mechanics experiment was used for auxiliary parameter validation. Firstly, the CT scanning system was used to scan the coal specimen, and the Otsu method was used to segment the original CT images to extract the three-dimensional fracture network and the distribution characteristics of mineral components. Secondly, the internal fracture and mineral distribution network model of the coal was reconstructed and imported into FLAC3D for modeling, and the numerical model was validated by fractal geometry method. The average error of fractal dimension between the original CT image and the corresponding numerical model is 1.422%, showing that the numerical model is basically consistent with the CT image of the coal, and the numerical model could reflect the distribution characteristics of the internal fractures and mineral components of the coal. Then, the basic mechanical parameters and acoustic emission (AE) response characteristics of CT reconstructed coal samples were obtained. And the mechanical parameters of the numerical model were calibrated. It can be concluded that the stress–strain curve of the coal is basically consistent with that of the numerical model, and the AE response characteristics of the coal are basically consistent with the evolution of the plastic zone of the numerical simulation, indicating that the numerical model and parameters are reasonable in this work. Moreover, the damage variable evolution of coal specimen in uniaxial compression and true triaxial loading was studied. Finally, the damage equation of coal under uniaxial compression and the relationship between damage variable and body strain energy under true triaxial loading were established.

Deep-Learning-Based Coronary Artery Calcium Detection from CT Image.

One of the most common methods for diagnosing coronary artery disease is the use of the coronary artery calcium score CT. However, the current diagnostic method using the coronary artery calcium score CT requires a considerable time, because the radiologist must manually check the CT images one-by-one, and check the exact range. In this paper, three CNN models are applied for 1200 normal cardiovascular CT images, and 1200 CT images in which calcium is present in the cardiovascular system. We conduct the experimental test by classifying the CT image data into the original coronary artery calcium score CT images containing the entire rib cage, the cardiac segmented images that cut out only the heart region, and cardiac cropped images that are created by using the cardiac images that are segmented into nine sub-parts and enlarged. As a result of the experimental test to determine the presence of calcium in a given CT image using Inception Resnet v2, VGG, and Resnet 50 models, the highest accuracy of 98.52% was obtained when cardiac cropped image data was applied using the Resnet 50 model. Therefore, in this paper, it is expected that through further research, both the simple presence of calcium and the automation of the calcium analysis score for each coronary artery calcium score CT will become possible.

Implementation of MR-Based Synthetic CT for Stereotactic Radiosurgery Dose Calculation

Most Gamma Knife stereotactic radiosurgery (GKRS) procedures have been performed using water-based dose calculation algorithms with magnetic resonance imaging (MRI) only. We present a dose calculation method that can take into account the inhomogeneity effect in routine GKRS treatment planning without CT imaging for every patient.Twenty-five patients who underwent GKRS at our institution with T1- and T2-weighted whole head MRI as well as brain CT images were selected for this study. Two set of synthetic computed tomographies (sCT) were generated from T1 and T2 MRI using cycle consistent generative adversial networks architecture (cGAN). Images from 20 patients were used to train the algorithm. The other 5 sets of data were used for testing and validation. The sCT images generated were evaluated in terms of the preservation of anatomical structures, the Hounsfield unit accuracy and the overall image quality. All sCT images from the 5 test patients were imported into GammaPlan version 11.1 for dosimetric analyses. We performed dose calculations for each test patient using the water-based TMR algorithm and the convolution algorithm with electron densities from the original CT, the T1-based sCT and the T2-based sCT.The main features of the original CT images (skull shape, bone & air inhomogeneity etc.) for the 5 test patients were successfully reproduced in the sCT images. sCT images from the T1-weighted MRI are in general closer to the original CT images compared to the T2-weighted MR images. For the same dose prescription and same shot arrangement, the maximum difference between the treatment shot times from the CT-based convolution calculation and the TMR calculation is 10.2%. The maximum differences between the treatment shot times from the CT-based convolution calculation and the convolution calculations from the sCTs from the T1 and T2 MR images are 2.3% and 3.8%, respectively.sCT images from brain MRI can be used for the inhomogeneity correction in GKRS dose calculation. T1-weighted MRI may work better than T2-weighted MRI for this purpose.F. Li: None. Y. Xu: None. O.D. Lemus: None. T.J. Wang: Research Grant; AbbVie, Merck, RTOG Foundation, Genentech. Honoraria; Elekta, Wolters Kluwer, Cancer Panels, Rutgers, University of Iowa. Consultant; Doximity, Elekta. Advisory Board; Novocure, AstraZeneca. Travel Expenses; AbbVie, Elekta, Novocure. Stock Options; Doximity. M.B. Sisti: None. C. Wuu: None.

A Lung Cancer Auxiliary Diagnostic Method: Deep Learning Based Mediastinal Lymphatic Partitions Segmentation for Cancer Staging

Accurate assessment of lymph node involvement plays an essential role in lung cancer treatment. The lymph node zonation map is widely used to depict the location of lymph node metastases, which could effectively reflect the degree of tumor invasion. According to the IASLC partitioning criteria and clinical work experience, a total of 15 partitions should be closely monitored. To facilitate doctors in their diagnostic work and improve their efficiency, we developed an automatic mediastinal lymph partitions segmentation algorithm in this study to provide direct visual guidance of the mediastinal area. However, the complex anatomical structure, ambiguous borders, similarity between targets, and variation between cases make the segmentation task more difficult.In this study, we propose a novel two-stage deep learning model for the automatic segmentation of all 15 mediastinal lymph partitions. U-Net is used as the backbone to extract spatial features to establish the mapping from original CT images to the segmentation mask in the first stage. Residual and dense connections are combined in the networks to construct to enhance the propagation of features and characterization effect of the models. An attention mechanism is designed which uses convolutional LSTM layers to control the attention in the feature transmission process to enable the model to focus more on the target area. To tackle the hard samples, which could not be segmented correctly only based on spatial features, we designed a convolutional recurrent network with LSTM structure to fuse features of different scales in the second stage to exploit temporal features to produce more granular segmentation results.A total of 80 CT data cases with different standards in some aspects are collected in this study and annotated by two experts based on the IASLC criteria. For the quantitative evaluation of our proposed method, an average Dice coefficient of 0.78 is achieved of fifteen partitions on the independent test set, which has exceeded the performance of widely used segmentation approaches and reached an advanced level. To further evaluate our method's approbation degree, a user study is conducted, which demonstrates the effectiveness of the designed method with a score of 4 out of 5.This study is the first to design and implement the automatic segmentation algorithm for mediastinal lymphatic partitions, facilitating doctors' work in diagnosis and treatment planning compared with manual identification. Qualitative and quantitative evaluations indicate the effectiveness and advancements of the algorithm.

Segmentation of the Upper Torso for Lung Cancer TTFields Treatment Planning

<h3>Purpose/Objective(s)</h3> Tumor Treating Fields (TTFields) is an FDA-approved treatment for glioblastoma multiforme (GBM) and malignant pleural mesothelioma (MPM). Moreover, TTFields therapy is currently investigated in a phase III clinical trial for the treatment of advanced Non-Small Cell Lung Cancer (NSCLC). Recent studies have shown that larger TTFields dose was associated with longer patients' survival. Therefore, personalized simulations to estimate the dose are performed as part of the patient treatment planning. For MPM and NSCLC treatment, these simulations require the segmentation of all upper torso tissues. A manual segmentation of the torso requires a few dozens of hours per patient and is impractical. Therefore, we have developed a computational method for semi-automatic segmentation of all upper torso tissues that are relevant to TTFields treatment planning. <h3>Materials/Methods</h3> We have incorporated a dataset of 40 CT images of NSCLC patients that underwent TTFields treatment in the lungs for this study. We have utilized threshold-based methods combined with morphological operations, region growing methods and known anatomical spatial relations to automatically identify and segment the lungs, bones, skin, muscle, fat, spinal cavity, costal cartilage, trachea and bronchi. Other structures such as the heart, blood vessels, liver, stomach, spleen, esophagus, diaphragm and intervertebral discs were semi-automatically segmented by using a few reference points that were provided by a human rater. Since there is no gold standard segmentation of the whole torso, an experienced radiation oncologist that is highly familiar with TTFields treatment inspected the results of the algorithm on top of the original CT images. <h3>Results</h3> The radiation oncologist has confirmed that the semi-automatic segmentation of the torso provides an adequate quality result for TTFields treatment planning for all cases. The segmentation time was reduced to one hour on a typical patient, compared to 20 hours that is the estimated time required for a fully manual segmentation. <h3>Conclusion</h3> We have presented a method for adequate quality segmentation of the upper torso in a reasonable time to facilitate TTFields treatment planning. In addition, this method facilitates the creation of a dataset for the development of state-of-the-art segmentation methods that utilize deep learning methods.

Chest Computerized Tomography Images under Iterative Model Reconstruction Algorithm in Patients with Lung Cancer

To explore the effect of the full iterative model reconstruction algorithm (IMR) on chest CT image processing and its adoption value in the clinical diagnosis of lung cancer patients, multislice spiral CT (MSCT) scans were performed on 96 patients with pulmonary nodules. Reconstruction was performed by hybrid iterative reconstruction (iDose4) and IMR2 algorithms. Then, the image contrast, spatial resolution, density resolution, image uniformity, and noise of the CT reconstructed image were recorded. The benign and malignant pulmonary nodules of patients were collected and classified into malignant pulmonary nodule group and benign pulmonary nodule group, and the differences in chest CT imaging characteristics between the two groups were compared. The subject’s receiver operating characteristic (ROC) curve was used to analyze the diagnostic sensitivity, specificity, and area under the curve (AUC) of CT for benign and malignant pulmonary nodules. It was found that the spatial resolution, density resolution, image uniformity, and contrast of the CT image reconstructed by the IMR2 algorithm were remarkably greater than those of the iDose4 algorithm, and the noise was considerably less than that of the iDose4 algorithm ( P &lt; 0.05 ). Among 96 patients with pulmonary nodules, 65 were malignant nodules, including 15 squamous cell carcinoma, 31 adenocarcinoma, and 19 small cell carcinomas. There were 31 cases of benign nodules, including 14 cases of hamartoma, 10 cases of tuberculous granuloma, 2 cases of sclerosing hemangioma, and 5 cases of diffuse lymphocyte proliferation. The pulmonary nodule malignant group and the pulmonary nodule benign group had statistical differences in pulmonary nodule size, nodule morphology, burr sign, lobular sign, vascular sign, bronchial sign, and pleural depression sign ( P &lt; 0.05 ). The sensitivity, specificity, and area under the curve (AUC) of IMR2 algorithm processing chest CT images for liver cancer diagnosis were 85.7%, 82.3%, and 0.815, respectively, which were significantly higher than the original CT images ( P &lt; 0.05 ). In short, chest MSCT based on the IMR2 algorithm can greatly improve the diagnosis efficiency of lung cancer and had practical significance for the timely detection of early lung cancer.

ACTIVE INVOLVEMENT OF RADIOLOGIST IN RADIATION ONCOLOGY PRACTICE

Advancements in Radiation Oncology from conventional to 3D conformal radiotherapy treatment demands expertise in many steps of radiation planning, the horizon of radiologist is now expanded by many folds and made radiologist as a integral part of the Radiation Oncology Department. A critical aspect of radiotherapy treatment planning (RTP) is determining how to deliver the required radiation dosage to cancer cells while minimising the exposure to normal tissue for which the prerequisite is identification and accurate delineation of tumour volume as well as normal structure resulted in an increase in the therapeutic ratio by reducing complication associated with normal tissue and allowing for higher target dosage and better local control. In modern radiotherapy CT images are the standard set of imaging modality required for the radiotherapy planning along with it many other modalities like MRI, PET or DSA are used by superimposing on original CT images in order to contour or delineate the structures defined by International Commission on Radiation Units and Measurements in Reports 50, 62 and 71 (ICRU) for radiotherapy planning which comprise of Gross tumour volume, clinical target volume, planning target volume, irradiated volume, Internal target volume and the normal structures as Organ at risk. It is self-evident that the contribution of a radiologist with a thorough knowledge of the development of these new modalities is critical for optimising the potential of these novel modes of radiation treatment delivery.

Real-time CT image generation based on voxel-by-voxel modeling of internal deformation by utilizing the displacement of fiducial markers.

To show the feasibility of real-time CT image generation technique utilizing internal fiducial markers that facilitate the evaluation of internal deformation. In the proposed method, a linear regression model that can derive internal deformation from the displacement of fiducial markers is built for each voxel in the training process before the treatment session. Marker displacement and internal deformation are derived from the four-dimensional computed tomography (4DCT) dataset. In the treatment session, the three-dimensional deformation vector field is derived according to the marker displacement, which is monitored by the real-time imaging system. The whole CT image can be synthesized by deforming the reference CT image with a deformation vector field in real-time. To show the feasibility of the technique, image synthesis accuracy and tumor localization accuracy were evaluated using the dataset generated by extended NURBS-Based Cardiac-Torso (XCAT) phantom and clinical 4DCT datasets from six patients, containing 10 CT datasets each. In the validation with XCAT phantom, motion range of the tumor in training data and validation data were about 10 and 15mm, respectively, so as to simulate motion variation between 4DCT acquisition and treatment session. In the validation with patient 4DCT dataset, eight CT datasets from the 4DCT dataset were used in the training process. Two excluded inhale CT datasets can be regarded as the datasets with large deformations more than training dataset. CT images were generated for each respiratory phase using the corresponding marker displacement. Root mean squared error (RMSE), normalized RMSE (NRMSE), and structural similarity index measure (SSIM) between the original CT images and the synthesized CT images were evaluated as the quantitative indices of the accuracy of image synthesis. The accuracy of tumor localization was also evaluated. In the validation with XCAT phantom, the mean NRMSE, SSIM, and three-dimensional tumor localization error were 7.5±1.1%, 0.95±0.02, and 0.4±0.3mm, respectively. In the validation with patient 4DCT dataset, the mean RMSE, NRMSE, SSIM, and three-dimensional tumor localization error in six patients were 73.7±19.6 HU, 9.2±2.6%, 0.88±0.04, and 0.8±0.6mm, respectively. These results suggest that the accuracy of the proposed technique is adequate when the respiratory motion is within the range of the training dataset. In the evaluation with a marker displacement larger than that of the training dataset, the mean RMSE, NRMSE, and tumor localization error were about 100 HU, 13%, and <2.0mm, respectively, except for one case having large motion variation. The performance of the proposed method was similar to those of previous studies. Processing time to generate the volumetric image was <100ms. We have shown the feasibility of the real-time CT image generation technique for volumetric imaging.

High through-plane resolution CT imaging with self-supervised deep learning

CT images for radiotherapy planning are usually acquired in thick slices to reduce the imaging dose, especially for pediatric patients, and to lessen the need for contouring and treatment planning on more slices. However, low through-plane resolution may degrade the accuracy of dose calculations. In this paper, a self-supervised deep learning workflow is proposed to synthesize high through-plane resolution CT images by learning from their high in-plane resolution features. The proposed workflow was designed to facilitate neural networks to learn the mapping from low-resolution (LR) to high-resolution (HR) images in the axial plane. During the inference step, the HR sagittal and coronal images were generated by feeding two parallel-trained neural networks with the respective LR sagittal and coronal images to the trained neural networks. The CT simulation images of a cohort of 75 patients with head and neck cancer (1 mm slice thickness) and 200 CT images of a cohort of 20 lung cancer patients (3 mm slice thickness) were retrospectively investigated in a cross-validation manner. The HR images generated with the proposed method were qualitatively (visual quality, image intensity profiles and preliminary observer study) and quantitatively (mean absolute error, edge keeping index, structural similarity index measurement, information fidelity criterion and visual information fidelity in pixel domain) inspected, while taking the original CT images of the head and neck and lung cancer patients as the reference. The qualitative results showed the capability of the proposed method for generating high through-plane resolution CT images with data from both groups of cancer patients. All the improvements in the measure metrics were confirmed to be statistically significant with paired two-sample t-test analysis. The innovative point of the work is that the proposed deep learning workflow for CT image generation with high through-plane resolution in radiotherapy is self-supervised, meaning that it does not rely on ground truth CT images to train the network. In addition, the assumption that the in-plane HR information can supervise the through-plane HR generation is confirmed. We hope that this will inspire more research on this topic to further improve the through-plane resolution of medical images.

Weakly-supervised progressive denoising with unpaired CT images.

Although low-dose CT imaging has attracted a great interest due to its reduced radiation risk to the patients, it suffers from severe and complex noise. Recent fully-supervised methods have shown impressive performances on CT denoising task. However, they require a huge amount of paired normal-dose and low-dose CT images, which is generally unavailable in real clinical practice. To address this problem, we propose a weakly-supervised denoising framework that generates paired original and noisier CT images from unpaired CT images using a physics-based noise model. Our denoising framework also includes a progressive denoising module that bypasses the challenges of mapping from low-dose to normal-dose CT images directly via progressively compensating the small noise gap. To quantitatively evaluate diagnostic image quality, we present the noise power spectrum and signal detection accuracy, which are well correlated with the visual inspection. The experimental results demonstrate that our method achieves remarkable performances, even superior to fully-supervised CT denoising with respect to the signal detectability. Moreover, our framework increases the flexibility in data collection, allowing us to utilize any unpaired data at any dose levels.