Computed Tomography Image Sets Research Articles

Dose distributions of proton therapy plans are robust against lowering the resolution of CTs combinedwith increasing noise.

Treatment planning in radiation therapy (RT) is performed on image sets acquired with commercial x-ray computed tomography (CT) scanners. Considering an increased frequency of verification scans for adaptive RT and the advent of alternatives to x-ray CTs, there is a need to review the requirements for image sets used in RT planning. This study aims to derive the required image quality (IQ) for the computation of the dose distribution in proton therapy (PT) regarding spatial resolution and the combination of spatial resolution and noise. The knowledge gained is used to explore the potential for dose reduction in tomography-guidedPT. Mathematical considerations indicate that the required spatial resolution for dose computation is on the scale of the set-up margins fed into the robust optimization. This hypothesis was tested by processing retrospectively 12 clinical PT cases, which reflect a variety of tumor localizations. Image sets were low-pass filtered and were made noisy in a generic manner. Dose distributions on the modified CT scans were computed with a Monte-Carlo dose engine. The similarity of these dose distributions with clinical ones was quantified with the gamma-index (1mm/1%). The potential reduction of the x-ray exposure compared to the planning CT scan was estimated. Dose distributions within the irradiated volume were robust against low-pass filtering of the CTs with kernels up to a full-width-at-half-maximum of 4mm, that is, the gamma pass rate (1mm/1%) was 98%. The limit of the filter width was 6mm for brain tumors and 8mm for targets in the abdomen. These pass rates remained approximately unchanged if a limited amount of noise was added to the CT image sets. The estimated potential reductions of the x-ray exposure were at least a factor of20. The requirements on IQ in terms of spatial resolution in combination with noise for computing the dose in PT are clearly lower than the IQ of current clinical planning. The results apply, for example, to ultra-low dose x-ray CTs, proton CTs with coarse spatial detection, and attenuation images from the joint reconstruction of time-of-flight PET scans.

CT-based artificial intelligence prediction model for ocular motility score of thyroid eye disease.

Thyroid eye disease (TED) is the most common orbital disease in adults. Ocular motility restriction is the primary complaint of patients, while its evaluation is quite difficult. The present study aimed to introduce an artificial intelligence (AI) model based on orbital computed tomography (CT) images for ocular motility score. A total of 410 sets of CT images and clinical data were obtained from the hospital. To build a triple classification predictive model for ocular motility score, multiple deep learning models were employed to extract features of images and clinical data. Subgroup analyses based on pertinent clinical features were performed to test the efficacy of models. The ResNet-34 network outperformed Alex-Net and VGG16-Net in prediction of ocular motility score, with the optimal accuracy (ACC) of 0.907, 0.870, and 0.890, respectively. Subgroup analyses indicated no significant difference in ACC between active or inactive phase, functional visual field diplopia or peripheral visual field diplopia (p > 0.05). However, in the gender subgroup, the prediction model performed more accurately in female patients than males (p = 0.02). In conclusion, the AI model based on CT images and clinical data successfully realized automatic scoring of ocular motility in TED patients. This approach potentially enhanced the efficiency and accuracy of ocular motility evaluation, thus facilitating clinical application.

A Spherical Cap Model of Epidural Hematomas.

Background Epidural hematomas (EDHs), which have a characteristic biconvex shape, are a type of post-traumatic intracranial mass. EDHs and other types of intracranial hematomas are often diagnosed with computed tomography (CT). The volumes of EDHs are important in treatment decisions and prognosis. Their volumes are usually estimated on CT using the "ABC" method, which is based on the ellipsoid shape rather than their biconvex shape. Objective To simulate the biconvex shape, we modeled the geometry of EDHs with two spherical caps. We aim to provide simpler estimation of EDH volumes in clinical settings, and eventually recommend a threshold for surgical evacuation. Methods Applying the relationship between the sphere radius, spherical cap height, and base circle radius, we derived formulas for the shape of an EDH, relating its largest diameter and location to the other two diameters. We also estimated EDH volumes using the spherical cap volume and conventional ABC formulas and then constructed a lookup table accordingly. Results Validation of the model was performed using 14 CT image sets from previously reported patients with EDHs. Our geometric model demonstrated accurate predictions. Themodel also allows reducing the number of parameters to be measured in the ABC method from three to one, the hematoma length, showcasing its potential as a reliable tool for clinical decision-making. Based on our model, an EDH longer than 7 cm would occupy more than 30 mL of the intracranial volume. Conclusion The proposed model offers a streamlined approach to estimating EDH volumes, reducing the complexity of parameters required for clinical assessments. We recommend a length of 7 cm as a threshold for surgical evacuation of EDHs. This acceleration in decision-making is crucial for managing critically injured patients with traumatic brain injuries. Further validation across diverse patient populations will enhance the generalizability and utility of this geometric modeling approach in clinical settings.

Robust explanation supervision for false positive reduction in pulmonary nodule detection.

Lung cancer is the deadliest and second most common cancer in the United States due to the lack of symptoms for early diagnosis. Pulmonary nodules are small abnormal regions that can be potentially correlated to the occurrence of lung cancer. Early detection of these nodules is critical because it can significantly improve the patient's survival rates. Thoracic thin-sliced computed tomography (CT) scanning has emerged as a widely used method for diagnosing and prognosis lung abnormalities. The standard clinical workflow of detecting pulmonary nodules relies on radiologists to analyze CT images to assess the risk factors of cancerous nodules. However, this approach can be error-prone due to the various nodule formation causes, such as pollutants and infections. Deep learning (DL) algorithms have recently demonstrated remarkable success in medical image classification and segmentation. As an ever more important assistant to radiologists in nodule detection, it is imperative ensure the DL algorithm and radiologist to better understand the decisions from each other. This study aims to develop a framework integrating explainable AI methods to achieve accurate pulmonary nodule detection. A robust and explainable detection (RXD) framework is proposed, focusing on reducing false positives in pulmonary nodule detection. Its implementation is based on an explanation supervision method, which uses nodule contours of radiologists as supervision signals to force the model to learn nodule morphologies, enabling improved learning ability on small dataset, and enable small dataset learning ability. In addition, two imputation methods are applied to the nodule region annotations to reduce the noise within human annotations and allow the model to have robust attributions that meet human expectations. The 480, 265, and 265 CT image sets from the public Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset are used for training, validation, and testing. Using only 10, 30, 50, and 100 training samples sequentially, our method constantly improves the classification performance and explanation quality of baseline in terms of Area Under the Curve (AUC) and Intersection over Union (IoU). In particular, our framework with a learnable imputation kernel improves IoU from baseline by 24.0% to 80.0%. A pre-defined Gaussian imputation kernel achieves an even greater improvement, from 38.4% to 118.8% from baseline. Compared to the baseline trained on 100 samples, our method shows less drop in AUC when trained on fewer samples. A comprehensive comparison of interpretability shows that our method aligns better with expert opinions. A pulmonary nodule detection framework was demonstrated using public thoracic CT image datasets. The framework integrates the robust explanation supervision (RES) technique to ensure the performance of nodule classification and morphology. The method can reduce the workload of radiologists and enable them to focus on the diagnosis and prognosis of the potential cancerous pulmonary nodules at the early stage to improve the outcomes for lung cancer patients.

Anatomical Variations between Sciatic Nerve and Piriformis Muscle on Computed Tomography Images from Radiotherapy Patients

Abstract Background: The sciatic nerve has several topographic variations, some of them in relation to the piriformis muscle. Such variations provide meaning insight while addressing pathologies related to sciatic nerve. Hence, this study was undertaken to check the anatomical variations of sciatic nerve using computed tomography (CT) images. Methodology: Thirty-three CT image series from pelvic radiotherapy patients were reviewed, with a DICOM viewer to assess the relation between the sciatic nerve and piriformis muscles, according to Beaton and Aston’s classification. Results: Thirty-three CT image sets and 66 gluteal regions were assessed. Type A of Beaton and Aston was found in 84.85% of the samples and Type B was found in 15.15%. No other types were found. Sixty-six percentage of the Type B classifications were bilateral. Conclusion: There is a 15.15% of topographical variation of sciatic nerve consistent with other studies. The use of medical images for this kind of study will allow more reliable results.

Deep learning image reconstruction algorithms in low-dose radiation abdominal computed tomography: assessment of image quality and lesion diagnostic confidence.

The image quality of computed tomography (CT) can be adversely affected by a low radiation dose, and reconstruction algorithms of an appropriate level may be useful in reducing this impact. Eight sets of CT images of a phantom were reconstructed with filtered back projection (FBP); adaptive statistical iterative reconstruction-Veo (ASiR-V) at 30% (AV-30), 50% (AV-50), 80% (AV-80), and 100% (AV-100); and deep learning image reconstruction (DLIR) at low (DL-L), medium (DL-M), and high (DL-H) levels. The noise power spectrum (NPS) and task transfer function (TTF) were measured. Thirty consecutive patients underwent low-dose radiation contrast-enhanced abdominal CT scans that were reconstructed using FBP, AV-30, AV-50, AV-80, and AV-100, and three levels of DLIR. The standard deviation (SD), signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR) of the hepatic parenchyma and paraspinal muscle were evaluated. Two radiologists assessed the subjective image quality and lesion diagnostic confidence using a 5-point Likert scale. In the phantom study, both a higher DLIR and ASiR-V strength and a higher radiation dose led less noise. The NPS peak and average spatial frequency of the DLIR algorithms were closer to those of FBP, as the tube current increased and declined as the level of ASiR-V and DLIR strengthened. The NPS average spatial frequency of DL-L were higher than those of AISR-V. In clinical studies, AV-30 demonstrated a higher SD and lower SNR and CNR compared to DL-M and DL-H (P<0.05). For qualitative assessment, DL-M produced the highest qualitative image quality scores, with the exception of overall image noise (P<0.05). The NPS peak, average spatial frequency, and SD were the highest and the SNR, CNR, and subjective scores were the lowest with FBP. Compared with FBP and ASiR-V, DLIR provided better image quality and noise texture both in the phantom and clinical studies, and DL-M maintained the best image quality and lesion diagnostic confidence in low-dose radiation abdominal CT.

Effectiveness of CT-image guidance in proton therapy for liver cancer and the importance of daily dose monitoring for tumors and organs at risk.

It is important to have precise image guidance throughout proton therapy in order to take advantage of the therapy's physical selectivity. We evaluated the effectiveness of computed tomography (CT)-image guidance in proton therapy for patients with hepatocellular carcinoma (HCC) by assessing daily proton dose distributions. The importance of daily CT image-guided registration and daily proton dose monitoring for tumors and organs at risk (OARs) was investigated. A retrospective analysis was conducted using 570 sets of daily CT (dCT) images throughout whole treatment fractions for 38 HCC patients who underwent passive scattering proton therapy with either a 66 cobalt gray equivalent (GyE)/10 fractions (n=19) or 76 GyE/20 fractions (n=19) protocol. The actual daily delivered dose distributions were estimated by forward calculation using the dCT sets, their corresponding treatment plans, and the recorded daily couch correction information. We then evaluated the daily changes of the dose indices D99% , V30GyE , and Dmax for the tumor volumes, non-tumorous liver, and other OARs, that is, stomach, esophagus, duodenum, colon, respectively. Contours were created for all dCT sets. We validated the efficacy of the dCT-based tumor registrations (hereafter, "tumor registration") by comparing them with the bone registration and diaphragm registration as a simulation of the treatment based on the positioning using the conventional kV X-ray imaging. The dose distributions and the indices of three registrations were obtained by simulation using the same dCT sets. In the 66 GyE/10 fractions, the daily D99% value in both the tumor and diaphragm registrations agreed with the planned value with 3%-6% (SD), and the V30GyE value for the liver agreed within ±3%; the indices in the bone registration showed greater deterioration. Nevertheless, tumor-dose deterioration occurred in all registration methods for two cases due to daily changes of body shape and respiratory condition. In the 76 GyE/20 fractions, in particular for such a treatment that the dose constraints for the OARs have to be cared in the original planning, the daily D99% in the tumor registration was superior to that in the other registration (p<0.001), indicating the effectiveness of the tumor registration. The dose constraints, set in the plan as the maximum dose for OARs (i.e., duodenum, stomach, colon, and esophagus) were maintained for 16 patients including seven treated with re-planning. For three patients, the daily Dmax increased gradually or changed randomly, resulting in an inter-fractional averaged Dmax higher than the constraints. The dose distribution would have been improved if re-planning had been conducted. The results of these retrospective analyses indicate the importance of daily dose monitoring followed by adaptive re-planning when needed. The tumor registration in proton treatment for HCC was effective to maintain the daily dose to the tumor and the dose constraints of OARs, particularly in the treatment where the maintenance for the dose constraints needs to be considered throughout the treatment. Nevertheless daily proton dose monitoring with daily CT imaging is important for more reliable and safer treatment.

Impact of Anatomical Position Errors on Dose Distribution in Head and Neck Radiotherapy and Robust Image Registration Against Anatomical Changes.

This study pursued two goals: Firstly, to search for anatomical structures strongly correlating with dose deterioration, and secondly to investigate the effectiveness of image registration focusing on critical anatomy by comparing it with a conventional method. The aim was to achieve robust image registration to correct for anatomical changes during treatment. Twenty patients with head and neck cancer were enrolled, and 68 simulation computed tomography (CT) and rescan CT image sets were retrospectively analyzed. Forty volumetric-modulated arc therapy and intensity-modulated proton therapy plans were generated and recalculated according to the rescan CT to evaluate the dose effects of anatomical changes. Correlation coefficients were calculated for the relationships between the six-axis motion of the anatomy and the dose indices for the clinical target volume (CTV) and organs at risk. In the image registration, we compared a conventional method and target-based registration that limited the registration range to the CTV and vertebrae. The CTV coverage and spinal cord dose were correlated with the position error associated with the pitch and vertical position of the vertebrae, and the parotid gland and oral cavity dose were strongly correlated with the position error associated with the roll of the clivus and mandible. The target registration improved CTV coverage and suppressed the increase in dose to organs at risk compared with conventional methods. Monitoring vertebral alignment, the assessment and correction of positioning errors associated with the clivus and mandible position errors are important to ensure the quality of daily treatment. Target-based registration may allow for more robust image registration.

A novel tool to evaluate and quantify radiation pneumonitis: A retrospective analysis of correlation of dosimetric parameters with volume of pneumonia patch.

In lung cancer, radiation-induced lung injury (RILI) or radiation pneumonitis (RP) are major concerns after radiotherapy. We investigated the correlation between volumes of RP lesions and their RP grades after radiotherapy. We retrospectively collected data from patients with non-small lung cancer that received curative doses to the thorax without undergoing chest radiotherapy before this treatment course. The post-treatment computed tomography (CT) image was used to register to the planning CT to evaluate the correlation between dosimetric parameters and volume of pneumonia patch by using deformable image registration. From January 1, 2019, to December 30, 2020, 71 patients with non-small cell lung cancer with 169 sets of CT images met our criteria for evaluation. In all patient groups, we found the RPv max and RP grade max to be significant (p<0.001). Some parameters that were related to the dose-volume histogram (DVH) and RP were lung Vx (x=1-66 Gy, percentage of lung volume received ≥x Gy), and mean lung dose. Comparing these parameters of the DVH with RP grade max showed that the mean lung dose and lung V1-V31 were significantly correlated. The cut-off point for the occurrence of symptoms in all patient groups, the RPv max value, was 4.79%, while the area under the curve was 0.779. In the groups with grades 1 and 2 RP, the dose curve of 26 Gy covered ≥80% of RP lesions in >80% of patients. Patients who had radiotherapy in combination with chemotherapy had significantly shorter locoregional progression-free survival (p=0.049) than patients who received radiation therapy in combination with target therapy. Patients with RPv max >4.79% demonstrated better OS (p=0.082). The percentage of RP lesion volume to total lung volume is a good indicator for quantifying RP. RP lesions can be projected onto the original radiation therapy plan using coverage of the 26 Gy isodose line to determine whether the lesion is RILI.

Automated Quantification of Pneumonia Infected Volume in Lung CT Images: A Comparison with Subjective Assessment of Radiologists.

To help improve radiologists' efficacy of disease diagnosis in reading computed tomography (CT) images, this study aims to investigate the feasibility of applying a modified deep learning (DL) method as a new strategy to automatically segment disease-infected regions and predict disease severity. We employed a public dataset acquired from 20 COVID-19 patients, which includes manually annotated lung and infections masks, to train a new ensembled DL model that combines five customized residual attention U-Net models to segment disease infected regions followed by a Feature Pyramid Network model to predict disease severity stage. To test the potential clinical utility of the new DL model, we conducted an observer comparison study. First, we collected another set of CT images acquired from 80 COVID-19 patients and process images using the new DL model. Second, we asked two chest radiologists to read images of each CT scan and report the estimated percentage of the disease-infected lung volume and disease severity level. Third, we also asked radiologists to rate acceptance of DL model-generated segmentation results using a 5-scale rating method. Data analysis results show that agreement of disease severity classification between the DL model and radiologists is >90% in 45 testing cases. Furthermore, >73% of cases received a high rating score (≥4) from two radiologists. This study demonstrates the feasibility of developing a new DL model to automatically segment disease-infected regions and quantitatively predict disease severity, which may help avoid tedious effort and inter-reader variability in subjective assessment of disease severity in future clinical practice.

Evaluation of edge detection algorithm of frontal image of facial contour in plastic surgery

With the improvement of medical levels and the continuous improvement of people’s living standards, the demand for beauty by the general public is increasing. The plastic surgery industry has also developed by leaps and bounds. People’s dissatisfaction with their own facial appearance, facial injuries and some other reasons have prompted people to carry out facial reconstruction, and facial plastic surgery has developed rapidly. However, in the current facial plastic surgery, the edge detection effect on the contour image is general. In order to improve the edge detection effect of facial contour lines in medical images, this paper proposed a facial contour line generation algorithm. First, the detection effects of four operators were compared. After comparing the effects, the Sobel operator was used as the input data to generate an edge detection algorithm. Then, the grayscale features of the tissue in the image and the symmetry of the image were used to perform bidirectional contour tracking on the detected image to extract facial contour lines. In addition, for facial contour features, the midpoint method can be used to generate auxiliary contours. The algorithm was verified by a set of facial CT (Computed Tomography) images in the experiment. The results showed that the new generation algorithm accelerated the edge detection speed, had good denoising performance, and enhanced the edge detection effect by about 12.05% compared with the traditional edge detection algorithm. The validity and practicability of facial edge detection were verified, and it provided a theoretical basis for further realizing the design of a facial contour digital image processing system.

RADIC:A tool for diagnosing COVID-19 from chest CT and X-ray scans using deep learning and quad-radiomics

Deep learning (DL) algorithms have demonstrated a high ability to perform speedy and accurate COVID-19 diagnosis utilizing computed tomography (CT) and X-Ray scans. The spatial information in these images was used to train DL models in the majority of relevant studies. However, training these models with images generated by radiomics approaches could enhance diagnostic accuracy. Furthermore, combining information from several radiomics approaches with time-frequency representations of the COVID-19 patterns can increase performance even further. This study introduces "RADIC", an automated tool that uses three DL models that are trained using radiomics-generated images to detect COVID-19. First, four radiomics approaches are used to analyze the original CT and X-ray images. Next, each of the three DL models is trained on a different set of radiomics, X-ray, and CT images. Then, for each DL model, deep features are obtained, and their dimensions are decreased using the Fast Walsh Hadamard Transform, yielding a time-frequency representation of the COVID-19 patterns. The tool then uses the discrete cosine transform to combine these deep features. Four classification models are then used to achieve classification. In order to validate the performance of RADIC, two benchmark datasets (CT and X-Ray) for COVID-19 are employed. The final accuracy attained using RADIC is 99.4% and 99% for the first and second datasets respectively. To prove the competing ability of RADIC, its performance is compared with related studies in the literature. The results reflect that RADIC achieve superior performance compared to other studies. The results of the proposed tool prove that a DL model can be trained more effectively with images generated by radiomics techniques than the original X-Ray and CT images. Besides, the incorporation of deep features extracted from DL models trained with multiple radiomics approaches will improve diagnostic accuracy.

Use of Chinese Visible Human Images to Enhance the Interpretation of CT or MRI Images: Application to Nasopharyngeal Structures Segmentation in the Treatment Planning System

Background: With leading morbidity among malignant tumors in otorhinolaryngology, Nasopharyngeal carcinoma (NPC) is one of the most frequent malignant tumors in China. Objectives: This study aimed to help radiotherapy doctors recognize and segment nasopharyngeal organs at risk of NPC and make a radiotherapy plan. Methods: The authors used B-spline and mutual information to transform, register, and fuse Chinese Visible Human images with the volunteer’s personalized computed tomography (CT) images, and integrated them into the Treatment Planning System (TPS). Consequently, Three-Dimensional Visualization Treatment Planning System (3DV+TPS) was created. To verify it, 3DV+TPS was deployed to identify and segment the nasopharyngeal organs at risk of NPC, and a questionnaire was filled out by radiotherapy doctors. Results: Results showed that 3DV+TPS can finish the registration and fusion of four sets of sectional anatomical images and individual CT images of volunteers in approximately 3 min and 50 sec. Conclusion: The registered and fused images can accurately reflect the position, outline, and adjacent space of the nasopharyngeal structure which is not clear in CT images. Therefore, it is helpful for recognizing and segmenting neural, muscular, and glandular structures. Through automatically registering and fusing color and CT gray images, 3DV+TPS improves the accuracy and efficiency of recognizing nasopharyngeal structures in making radiotherapy plans. It is also useful to improve the teaching quality of tumor radiotherapy for medical students and interns as well.

High-Resolution Three-Dimensional Hybrid MRI + Low Dose CT Vocal Tract Modeling: A Cadaveric Pilot Study

<h2>Summary</h2><h3>Objectives</h3> MRI based vocal tract models have many applications in voice research and education. These models do not adequately capture bony structures (e.g. teeth, mandible), and spatial resolution is often relatively low in order to minimize scanning time. Most MRI sequences achieve 3D vocal tract coverage at gross resolutions of 2 mm³ within a scan time of <20 seconds. Computed tomography (CT) is well suited for vocal tract imaging, but is infrequently used due to the risk of ionizing radiation. In this cadaveric study, a single, extremely low-dose CT scan of the bony structures is blended with accelerated high-resolution (1 mm³) MRI scans of the soft tissues, creating a high-resolution hybrid CT-MRI vocal tract model. <h3>Methods</h3> Minimum CT dosages were determined and a custom 16-channel airway receiver coil for accelerated high (1 mm³) resolution MRI was evaluated. A rigid body landmark based partial volume registration scheme was then applied to the images, creating a hybrid CT-MRI model that was segmented in Slicer. <h3>Results</h3> Ultra-low dose CT produced images with sufficient quality to clearly visualize the bone, and exposed the cadaver to 0.06 mSv. This is comparable to atmospheric exposures during a round trip transatlantic flight. The custom 16-channel vocal tract coil produced acceptable image quality at 1 mm³ resolution when reconstructed from ∼6 fold undersampled data. High (1 mm³) resolution MR imaging of short (<10 seconds) sustained sounds was achieved. The feasibility of hybrid CT-MRI vocal tract modeling was successfully demonstrated using the rigid body landmark based partial volume registration scheme. Segmentations of CT and hybrid CT-MRI images provided more detailed 3D representations of the vocal tract than 2 mm³ MRI based segmentations. <h3>Conclusions</h3> The method described in this study indicates that high-resolution CT and MR image sets can be combined so that structures such as teeth and bone are accurately represented in vocal tract reconstructions. Such scans will aid learning and deepen understanding of anatomical features that relate to voice production, as well as furthering knowledge of the static and dynamic functioning of individual structures relating to voice production.

Development of an artificial intelligence-assisted computed tomography diagnosis technology for rib fracture and evaluation of its clinical usefulness

Artificial intelligence algorithms utilizing deep learning are helpful tools for diagnostic imaging. A deep learning-based automatic detection algorithm was developed for rib fractures on computed tomography (CT) images of high-energy trauma patients. In this study, the clinical effectiveness of this algorithm was evaluated. A total of 56 cases were retrospectively examined, including 46 rib fractures and 10 control cases from our hospital, between January and June 2019. Two radiologists annotated the fracture lesions (complete or incomplete) for each CT image, which is considered the “ground truth.” Thereafter, the algorithm’s diagnostic results for all cases were compared with the ground truth, and the sensitivity and number of false positive (FP) results per case were assessed. The radiologists identified 199 images with a fracture. The sensitivity of the algorithm was 89.8%, and the number of FPs per case was 2.5. After additional learning, the sensitivity increased to 93.5%, and the number of FPs was 1.9 per case. FP results were found in the trabecular bone with the appearance of fracture, vascular grooves, and artifacts. The sensitivity of the algorithm used in this study was sufficient to aid the rapid detection of rib fractures within the evaluated validation set of CT images.

Approaching automated applicator digitization from a new angle: Using sagittal images to improve deep learning accuracy and robustness in high-dose-rate prostate brachytherapy

PURPOSETo automate the segmentation of treatment applicators on computed tomography (CT) images for high-dose-rate (HDR) brachytherapy prostate patients implanted with titanium needles with the goals of improving plan quality and reducing the patient's time under anesthesia. METHODSThe investigation was performed using 57 retrospective, interstitial prostate treatments randomly assigned to training (n = 27), validation (n = 10), and testing (n = 20). Unique to this work, the CT image set was reformatted into 2D sagittal slices instead of the default axial orientation. A deep learning-based segmentation was performed using a 2D U-Net architecture followed by a density-based linkage clustering algorithm to classify individual catheters in 3D. Potential confounders, such as gold seeds and conjoined applicators with intersecting needle geometries, were corrected using a customized polynomial fitting algorithm. The geometric agreement of the automated digitization was evaluated against the clinically treated manual digitization to measure tip and shaft errors in the reconstruction. RESULTSThe proposed algorithm achieved tip and shaft agreements of -0.1 ± 0.6 mm (range -1.8 mm to 1.4 mm) and 0.13 ± 0.09 mm (maximum 0.96 mm), respectively on a data set with 20 patients and 353 total needles. Our method was able to separate all intersecting applicators reliably. The time to generate the automated applicator digitization averaged approximately 1 min. CONCLUSIONSUsing sagittal instead of axial images for 2D segmentation of interstitial brachytherapy applicators produced submillimeter agreement with manual segmentation. The automated digitization of interstitial applicators in prostate brachytherapy has the potential to improve quality and consistency while reducing the patient's time under anesthesia.

Applying Quantitative Radiographic Image Markers to Predict Clinical Complications After Aneurysmal Subarachnoid Hemorrhage: A Pilot Study.

Accurately predicting clinical outcome of aneurysmal subarachnoid hemorrhage (aSAH) patients is difficult. The purpose of this study was to develop and test a new fully-automated computer-aided detection (CAD) scheme of brain computed tomography (CT) images to predict prognosis of aSAH patients. A retrospective dataset of 59 aSAH patients was assembled. Each patient had 2 sets of CT images acquired at admission and prior-to-discharge. CAD scheme was applied to segment intracranial brain regions into four subregions, namely, cerebrospinal fluid (CSF), white matter (WM), gray matter (GM), and leaked extraparenchymal blood (EPB), respectively. CAD then detects sulci and computes 9 image features related to 5 volumes of the segmented sulci, EPB, CSF, WM, and GM and 4 volumetrical ratios to sulci. Subsequently, applying a leave-one-case-out cross-validation method embedded with a principal component analysis (PCA) algorithm to generate optimal feature vector, 16 support vector machine (SVM) models were built using CT images acquired either at admission or prior-to-discharge to predict each of eight clinically relevant parameters commonly used to assess patients' prognosis. Finally, a receiver operating characteristics (ROC) method was used to evaluate SVM model performance. Areas under ROC curves of 16 SVM models range from 0.62 ± 0.07 to 0.86 ± 0.07. In general, SVM models trained using CT images acquired at admission yielded higher accuracy to predict short-term clinical outcomes, while SVM models trained using CT images acquired prior-to-discharge demonstrated higher accuracy in predicting long-term clinical outcomes. This study demonstrates feasibility to predict prognosis of aSAH patients using new quantitative image markers generated by SVM models.

Deep Learning Model With Convolutional Neural Network for Detecting and Segmenting Hepatocellular Carcinoma in CT: A Preliminary Study.

IntroductionHepatocellular carcinoma (HCC) is one of the most common malignancies in the world. Early detection and accurate diagnosis of HCC play an important role in patient management. This study aimed to develop a convolutional neural network-based model to identify and segment HCC lesions utilizing dynamic contrast agent-enhanced computed tomography (CT).MethodsThis retrospective study used CT image sets of histopathology-confirmed hepatocellular carcinoma over three phases (arterial, venous, and delayed). The proposed convolutional neural network (CNN) segmentation method was based on the U-Net architecture and trained using the domain adaptation technique. The proposed method was evaluated using 115 liver masses of 110 patients (87 men and 23 women; mean age, 56.9 years ± 11.9 (SD); mean mass size, 6.0 cm ± 3.6). The sensitivity for identifying HCC of the model and Dice score for segmentation of liver masses between radiologists and the CNN model were calculated for the test set.ResultsThe sensitivity for HCC identification of the model was 100%. The median Dice score for HCC segmenting between radiologists and the CNN model was 0.81 for the test set.ConclusionDeep learning with CNN had high performance in the identification and segmentation of HCC on dynamic CT.

Real-time virtual intraoperative CT in endoscopic sinus surgery.

Endoscopic sinus surgery (ESS) is typically guided under preoperative computed tomography (CT), which increasingly diverges from actual patient anatomy as the surgery progresses. Studies have reported that the revision surgery rate in ESS ranges between 28 and 47%. This paper presents a method that can update the preoperative CT in real time to improve surgical completeness in ESS. The work presents and compares three novel methods that use instrument motion data and anatomical structures to predict surgical modifications in real time. The methods use learning techniques, such as nonparametric filtering and Gaussian process regression, to correlate surgical modifications with instrument tip positions, tip trajectories, and instrument shapes. Preoperative CT image sets are updated with modification predictions to serve as a virtual intraoperative CT. The three methods were compared in eight ESS cadaver cases, which were performed by five surgeons and included the following representative ESS operations: maxillary antrostomy, uncinectomy, anterior and posterior ethmoidectomy, and sphenoidotomy. Experimental results showed accuracy metrics that were clinically acceptable with dice similarity coefficients > 86%, with F-score > 92% and precision > 89.91% in surgical completeness evaluation. Among the three methods, the tip trajectory-based estimator had the highest precision of 96.87%. This work demonstrated that virtually modified intraoperative CT scans improved the consistency between the actual surgical scene and the reference model, and could lead to improved surgical completeness in ESS. Compared to actual intraoperative CT scans, the proposed method has no impact on existing surgical protocols, does not require extra hardware, does not expose the patient to radiation, and does not lengthen time under anesthesia.

Deep Learning Based Identification and Segmentation of Lung Tumors on Computed Tomography Images

Rapid and accurate estimation of tumor burden in biomedical images is essential for precisely monitoring cancer progression and assessing therapeutic response. The ability to detect and segment tumors using an automated approach is a key part of this task. Despite recent advances from deep learning, lung tumor delineation remains challenging, particularly when the tumor bounding box is not provided to the model. We hypothesized that clinical radiation oncology contours could supply a large enough dataset of 3D tumor segmentations to enable more accurate models. We developed and validated a deep learning-based model to identify and segment primary and metastatic lung tumors on computed tomography (CT) images.We curated a dataset consisting of CT images and clinical segmentations of 1,916 lung tumors in 1,504 patients who received radiation treatment for one or more primary or metastatic lung tumors. Segmentation quality was independently verified by a radiation oncologist using a custom web application. This dataset was used to train two 3D U-Net convolutional neural networks with varying model properties: one using high-resolution and small input volumes, and one using low-resolution and large input volumes. Models were ensembled together during validation. Performance was evaluated using an external held-out test set of CT images and segmentations from 59 patients with a single primary or metastatic lung tumor, treated at a separate clinical site. This test set consisted of 50 primary lung cancers and 9 metastases. To benchmark model performance against physicians, the test set was also contoured by two additional radiation oncologists.Median tumor volume in the external test set was 80.48 cubic centimeters (interquartile range [IQR]: 14.40 to 177.65). The segmentations generated by the ensembled model produced a mean Dice coefficient of 0.62 (IQR: 0.47 to 0.85) on the test set. The sensitivity for detecting a tumor, as defined by correctly predicting at least one voxel within a ground truth tumor, was 93.2%, and the Dice coefficient for the scans with correctly identified lesions was 0.67 (IQR: 0.53 to 0.85). In comparison, the mean interobserver Dice coefficient for the three physicians on the test set was 0.76 (IQR: 0.70 to 0.84). We observed strong correlation between physician-determined tumor size and model-predicted tumor size (Pearson correlation, r = 0.69, P < 0.0001).An end-to-end deep learning-based model was able to identify and segment lung tumors in a completely automated fashion, with near-expert level performance. Such models could soon be useful for clinical contouring and automatic quantification of tumor burden.