Predicting Perceived Reporting Complexity of Abdominopelvic Computed Tomography With Deep Learning.

OBJECTIVE: The purpose of this pilot study was to examine human and automated estimates of reporting complexity for computed tomography (CT) studies of the abdomen and pelvis. METHODS: A total of 1019 CT studies were reviewed and categorized into 3 complexity categories by 3 abdominal radiologists, and the majority classification was used as ground truth. Studies were randomized into a training set of 498 studies and a test set of 521 studies. A 2-stage neural network model was trained on the training set; the first-stage image-level classifier produces image embeddings that are used in the second-stage sequential model to provide a study-level prediction. RESULTS: All 3 human reviewers agreed on ratings for 470 of the 1019 studies (46%); at least 2 of the 3 reviewers agreed on ratings for 1010 studies (99%). After training, the neural network model predicted complexity labels that agreed with the radiologist consensus rating on 55% of the studies; 90% of the incorrect predicted categories were errors where the predicted category differed from the consensus rating by one level of complexity. CONCLUSIONS: There is moderate interrater agreement in radiologist-perceived reporting complexity for CT studies of the abdomen and pelvis. Automated prediction of reporting complexity in radiology studies may be a useful adjunct to radiology practice analytics.

Deep Learning: An Update for Radiologists.

Deep learning is a class of machine learning methods that has been successful in computer vision. Unlike traditional machine learning methods that require hand-engineered feature extraction from input images, deep learning methods learn the image features by which to classify data. Convolutional neural networks (CNNs), the core of deep learning methods for imaging, are multilayered artificial neural networks with weighted connections between neurons that are iteratively adjusted through repeated exposure to training data. These networks have numerous applications in radiology, particularly in image classification, object detection, semantic segmentation, and instance segmentation. The authors provide an update on a recent primer on deep learning for radiologists, and they review terminology, data requirements, and recent trends in the design of CNNs; illustrate building blocks and architectures adapted to computer vision tasks, including generative architectures; and discuss training and validation, performance metrics, visualization, and future directions. Familiarity with the key concepts described will help radiologists understand advances of deep learning in medical imaging and facilitate clinical adoption of these techniques. Online supplemental material is available for this article. ©RSNA, 2021.

Artificial Intelligence for Medical Image Analysis: A Guide for Authors and Reviewers.

OBJECTIVE: The purpose of this article is to highlight best practices for writing and reviewing articles on artificial intelligence for medical image analysis. CONCLUSION: Artificial intelligence is in the early phases of application to medical imaging, and patient safety demands a commitment to sound methods and avoidance of rhetorical and overly optimistic claims. Adherence to best practices should elevate the quality of articles submitted to and published by clinical journals.

Refining Convolutional Neural Network Detection of Small-Bowel Obstruction in Conventional Radiography.

OBJECTIVE: The purpose of this study was to evaluate improvement of convolutional neural network detection of high-grade small-bowel obstruction on conventional radiographs with increased training set size. MATERIALS AND METHODS: A set of 2210 abdominal radiographs from one institution (image set 1) had been previously classified into obstructive and nonobstructive categories by consensus judgments of three abdominal radiologists. The images were used to fine-tune an initial convolutional neural network classifier (stage 1). An additional set of 13,935 clinical images from the same institution was reduced to 5558 radiographs (image set 2) primarily by retaining only images classified positive for bowel obstruction by the initial classifier. These images were classified into obstructive and nonobstructive categories by an abdominal radiologist. The combined 7768 radiographs were used to train additional classifiers (stage 2 training). The best classifiers from stage 1 and stage 2 training were evaluated on a held-out test set of 1453 abdominal radiographs from image set 1. RESULTS: The ROC AUC for the neural network trained on image set 1 was 0.803; after stage 2, the ROC AUC of the best model was 0.971. By use of an operating point based on maximizing the validation set Youden J index, the stage 2-trained model had a test set sensitivity of 91.4% and specificity of 91.9%. Classification performance increased with training set size, reaching a plateau with over 200 positive training examples. CONCLUSION: Accuracy of detection of high-grade small-bowel obstruction with a convolutional neural network improves significantly with the number of positive training radiographs.

Tackling the Radiological Society of North America Pneumonia Detection Challenge.

OBJECTIVE. We provide overviews of deep learning approaches used by two top-placing teams for the 2018 Radiological Society of North America (RSNA) Pneumonia Detection Challenge. CONCLUSION. Practical applications of deep learning techniques, as well as insights into the annotation of the data, were keys to success in accurately detecting pneumonia on chest radiographs for the competition.

Detection of Traumatic Pediatric Elbow Joint Effusion Using a Deep Convolutional Neural Network.

OBJECTIVE: The purpose of this study is to determine whether a deep convolutional neural network (DCNN) trained on a dataset of limited size can accurately diagnose traumatic pediatric elbow effusion on lateral radiographs. MATERIALS AND METHODS: A total of 901 lateral elbow radiographs from 882 pediatric patients who presented to the emergency department with upper extremity trauma were divided into a training set (657 images), a validation set (115 images), and an independent test set (129 images). The training set was used to train DCNNs of varying depth, architecture, and parameter initialization, some trained from randomly initialized parameter weights and others trained using parameter weights derived from pretraining on an ImageNet dataset. Hyperparameters were optimized using the validation set, and the DCNN with the highest ROC AUC on the validation set was selected for further performance testing on the test set. RESULTS: The final trained DCNN model had an ROC AUC of 0.985 (95% CI, 0.966-1.000) on the validation set and 0.943 (95% CI, 0.884-1.000) on the test set. On the test set, sensitivity was 0.909 (95% CI, 0.788-1.000), specificity was 0.906 (95% CI, 0.844-0.958), and accuracy was 0.907 (95% CI, 0.843-0.951). CONCLUSION: Accurate diagnosis of traumatic pediatric elbow joint effusion can be achieved using a DCNN.

Deep Learning: A Primer for Radiologists.

Deep learning is a class of machine learning methods that are gaining success and attracting interest in many domains, including computer vision, speech recognition, natural language processing, and playing games. Deep learning methods produce a mapping from raw inputs to desired outputs (eg, image classes). Unlike traditional machine learning methods, which require hand-engineered feature extraction from inputs, deep learning methods learn these features directly from data. With the advent of large datasets and increased computing power, these methods can produce models with exceptional performance. These models are multilayer artificial neural networks, loosely inspired by biologic neural systems. Weighted connections between nodes (neurons) in the network are iteratively adjusted based on example pairs of inputs and target outputs by back-propagating a corrective error signal through the network. For computer vision tasks, convolutional neural networks (CNNs) have proven to be effective. Recently, several clinical applications of CNNs have been proposed and studied in radiology for classification, detection, and segmentation tasks. This article reviews the key concepts of deep learning for clinical radiologists, discusses technical requirements, describes emerging applications in clinical radiology, and outlines limitations and future directions in this field. Radiologists should become familiar with the principles and potential applications of deep learning in medical imaging. ©RSNA, 2017.

Transfer Learning with Convolutional Neural Networks for Classification of Abdominal Ultrasound Images.

The purpose of this study is to evaluate transfer learning with deep convolutional neural networks for the classification of abdominal ultrasound images. Grayscale images from 185 consecutive clinical abdominal ultrasound studies were categorized into 11 categories based on the text annotation specified by the technologist for the image. Cropped images were rescaled to 256 × 256 resolution and randomized, with 4094 images from 136 studies constituting the training set, and 1423 images from 49 studies constituting the test set. The fully connected layers of two convolutional neural networks based on CaffeNet and VGGNet, previously trained on the 2012 Large Scale Visual Recognition Challenge data set, were retrained on the training set. Weights in the convolutional layers of each network were frozen to serve as fixed feature extractors. Accuracy on the test set was evaluated for each network. A radiologist experienced in abdominal ultrasound also independently classified the images in the test set into the same 11 categories. The CaffeNet network classified 77.3% of the test set images accurately (1100/1423 images), with a top-2 accuracy of 90.4% (1287/1423 images). The larger VGGNet network classified 77.9% of the test set accurately (1109/1423 images), with a top-2 accuracy of VGGNet was 89.7% (1276/1423 images). The radiologist classified 71.7% of the test set images correctly (1020/1423 images). The differences in classification accuracies between both neural networks and the radiologist were statistically significant (p < 0.001). The results demonstrate that transfer learning with convolutional neural networks may be used to construct effective classifiers for abdominal ultrasound images.

Histogram-Based Discrimination of Intravenous Contrast in Abdominopelvic Computed Tomography.

OBJECTIVE: The aim of this study was to evaluate the accuracy of fully automated machine learning methods for detecting intravenous contrast in computed tomography (CT) studies of the abdomen and pelvis. METHODS: A set of 591 labeled CT image volumes of the abdomen and pelvis was obtained from 5 different CT scanners, of which 434 (73%) were performed with intravenous contrast. A stratified split of this set was performed into training and test sets of 443 and 148 studies, respectively. Subsequently, support vector machine and logistic regression classifiers were trained using 5-fold cross-validation for parameter optimization. RESULTS: The best in-sample performance was seen with a support vector machine classifier with a χ kernel (98.9% accuracy); however, test set performance was similar across the trained classifiers, with 95% to 97% accuracy. CONCLUSIONS: Histogram-based automated classifiers for the presence of intravenous contrast are accurate and may be useful for verifying the accurate labeling of the presence of intravenous contrast in CT body studies.

Patient Vertical Centering and Correlation with Radiation Output in Adult Abdominopelvic CT.

The purpose of this study was to determine if there is a significant effect, independent of patient size, of patient vertical centering on the current-modulated CT scanner radiation output in adult abdominopelvic CT. A phantom was used to evaluate calculation of vertical positioning and effective diameter at five different table heights. In addition, 656 consecutive contrast-enhanced abdominopelvic scans using the same protocol and automatic tube current modulation settings on a Philips Brilliance 64 MDCT scanner were retrospectively evaluated. The vertical position of the patient center of mass and the average effective diameter of the scanned patient were computed using the reconstructed images. The average volume CT dose index (CTDIvol) for each scan was recorded. The mean patient center of mass y coordinate ranged from -3.7 to 6.7 cm (mean ± SD, 2.8 ± 1.2 cm), indicating that patients were on average positioned slightly below the scanner isocenter. There was a slight tendency for smaller patients to be mis-centered lower than larger patients. Average CTDIvol closely fit a quadratic regression curve with respect to mean effective diameter. However, the value of the regression coefficient relating CTDIvol to the patient's vertical position was nearly zero, indicating only a very slight increase in CTDIvol with patient mis-centering for the scanner used in this study. The techniques used here may be useful both for automated evaluation of proper patient positioning in CT and for estimating the radiation dose effects of patient mis-centering for any CT scanner.

Quantitative assessment of solid renal masses by contrast-enhanced ultrasound with time-intensity curves: how we do it.

PURPOSE: To discuss the evaluation of the enhancement curve over time of the major renal cell carcinoma (RCC) subtypes, oncocytoma, and lipid-poor angiomyolipoma, to aid in the preoperative differentiation of these entities. Differentiation of these lesions is important, given the different prognoses of the subtypes, as well as the desire to avoid resecting benign lesions. METHODS: We discuss findings from CT, MR, and US, but with a special emphasis on contrast-enhanced ultrasound (CEUS). CEUS technique is described, as well as time-intensity curve analysis. RESULTS: Examples of each of the major RCC subtypes (clear cell, papillary, and chromophobe) are shown, as well as examples of oncocytoma and lipid-poor angiomyolipoma. For each lesion, the time-intensity curve of enhancement on CEUS is reviewed, and correlated with the enhancement curve over time reported for multiphase CT and MR. CONCLUSIONS: Preoperative differentiation of the most common solid renal masses is important, and the time-intensity curves of these lesions show some distinguishing features that can aid in this differentiation. The use of CEUS is increasing, and as a modality it is especially well suited to the evaluation of the time-intensity curve.

Pulmonary pseudoemboli: a new artifact arising from a commercial metal artifact reduction algorithm for computed tomographic image reconstruction.

Metal artifact reduction algorithms for computed tomographic (CT) image reconstruction have recently become commercially available on modern CT scanners for reducing artifacts from orthopedic hardware. However, we have observed that a commercial orthopedic metal artifact reduction algorithm can produce the appearance of artifactual pulmonary emboli when applied to spinal hardware in contrast-enhanced CT scans of the chest. We provide 4 case examples demonstrating this previously undescribed artifact.

Complete tumor encapsulation on magnetic resonance imaging: a potentially useful imaging biomarker for better survival in solitary large hepatocellular carcinoma.

The aim of this study was to determine the prognostic value of complete tumor encapsulation as visualized on magnetic resonance imaging (MRI) in patients with a solitary large hepatocellular carcinoma (HCC) beyond the Milan criteria for liver transplantation (LT). Between December 2000 and March 2011, 57 patients who had a solitary HCC exceeding 5 cm in diameter at the time of initial MRI before any treatment were identified. MRI images of the patients were independently reviewed by 2 experienced readers for the presence of complete tumoral encapsulation. The medical records of the patients were reviewed for an outcome analysis. Thirty of the 57 patients had completely encapsulated HCC according to MRI. There was excellent interobserver agreement between the 2 readers for the assessment of complete encapsulation (κ=0.86). Overall survival was significantly longer for patients with completely encapsulated HCC versus patients with incompletely or nonencapsulated tumors (P<0.001), and this included a subanalysis of 33 patients who received locoregional treatment (LRT; P=0.04). The presence of complete encapsulation was a strong predictor for survival in these patients according to both univariate [hazard ratio (HR)=0.24, 95% confidence interval (CI)=0.12-0.52, P<0.001] and multivariate analyses (HR=0.25, 95% CI=0.07-0.85, P=0.03). The rates of down-staging (P<0.001) and eventual LT (P=0.02) after LRT were also significantly higher in the patients with completely encapsulated tumors. In conclusion, complete tumor encapsulation on MRI is a potentially useful predictor for favorable biology in patients with a solitary large HCC. This new imaging biomarker may have a role in treatment selection for patients whose tumors exceed the Milan criteria size limits.

Automated estimation of abdominal effective diameter for body size normalization of CT dose.

Most CT dose data aggregation methods do not currently adjust dose values for patient size. This work proposes a simple heuristic for reliably computing an effective diameter of a patient from an abdominal CT image. Evaluation of this method on 106 patients scanned on Philips Brilliance 64 and Brilliance Big Bore scanners demonstrates close correspondence between computed and manually measured patient effective diameters, with a mean absolute error of 1.0 cm (error range +2.2 to -0.4 cm). This level of correspondence was also demonstrated for 60 patients on Siemens, General Electric, and Toshiba scanners. A calculated effective diameter in the middle slice of an abdominal CT study was found to be a close approximation of the mean calculated effective diameter for the study, with a mean absolute error of approximately 1.0 cm (error range +3.5 to -2.2 cm). Furthermore, the mean absolute error for an adjusted mean volume computed tomography dose index (CTDIvol) using a mid-study calculated effective diameter, versus a mean per-slice adjusted CTDIvol based on the calculated effective diameter of each slice, was 0.59 mGy (error range 1.64 to -3.12 mGy). These results are used to calculate approximate normalized dose length product values in an abdominal CT dose database of 12,506 studies.

Automated pediatric abdominal effective diameter measurements versus age-predicted body size for normalization of CT dose.

There has been increasing interest in adjusting CT radiation dose data for patient body size. A method for automated computation of the abdominal effective diameter of a patient from a CT image has previously only been tested in adult patients. In this work, we tested the method on a set of 128 pediatric patients aged 0.8 to 12.9 years (average 8.0 years, SD = 3.7 years) who had CT abdomen/pelvis exams performed on a Toshiba Aquilion 64 scanner. For this set of patients, age-predicted abdominal effective diameter extrapolated based on data from the International Commission on Radiation Units and Measurements was a relatively poor predictor of measured effective diameter. The mean absolute percentage error between the CTDI normalization coefficient calculated from a manually measured effective diameter and the coefficient determined by age-predicted effective diameter was 12.3 % with respect to a 32 cm phantom (range 0.0-52.8 %, SD 8.7 %) and 12.9 % with respect to a 16 cm phantom (range 0.0-56.4 %, SD 9.2 %). In contrast, there is a close correspondence between the automated and manually measured patient effective diameters, with a mean absolute error of 0.6 cm (error range 0.2-1.3 cm). This correspondence translates into a high degree of correspondence between normalization coefficients determined by automated and manual measurements; the mean absolute percentage error was 2.1 % with respect to a 32 cm phantom (range 0.0-8.1 %, SD = 1.4 %) and 2.3 % with respect to a 16 cm phantom (range 0.0-9.3 %, SD = 1.6 %).

Feminizing Adrenocortical Tumor with Multiple Recurrences: A Case Report.

Feminizing adrenocortical tumors (FATs) are exceptionally rare primary adrenal neoplasms that cause high estrogen and low testosterone levels. They are most common in adult males, typically presenting with gynecomastia, hypogonadism, and weight loss. They are almost always malignant, with a poor prognosis and a high recurrence rate. We report a case of a 35-year-old man with an adrenal FAT with high estrogen (181 pg/mL) and low testosterone (37 ng/dL) who presented with gynecomastia, erectile dysfunction, subclinical Cushing syndrome, and pain localizing to different regions of the torso. There was no evidence of metastatic disease initially as seen by visualization of a well-marginated mass on computed tomography scan. Surgical resection of the FAT was performed, and the mass was confirmed to be a low-grade tumor. Clinical symptoms were resolved after surgery. Despite complete resection with negative margins, the patient subsequently had two separate local metastatic recurrences within a few years, treated with a combination of further surgery and medical intervention. This case highlights the unique features of an exceedingly rare adrenal tumor and stresses the importance of early detection and vigilant surveillance following resection due to high recurrence rates.

Perinephric myxoid pseudotumor of fat: a multimodality imaging case series.

Perinephric myxoid pseudotumor of fat (PMPF) is an unusual clinical entity with few prior imaging case reports. We report a multimodality imaging case series of PMPF, consisting of four cases seen in our department with both imaging studies and histopathologic confirmation. Three of the four patients had a history of advanced non-neoplastic renal disease. The perirenal masses in these cases varied in size and appearance. Some lesions resembled cysts or contained macroscopic fat. Enhancement was equivocal on CT, but prominent in one case on MRI and in another on contrast-enhanced ultrasound. Although not known to be malignant, PMPF may be confused for a cyst, liposarcoma, or hypovascular solid neoplasm on imaging. The dominant mass was resected in two cases because of concern for malignancy, while percutaneous CT-guided biopsy was performed in the other two. Mouse double minute 2 (MDM2) gene amplification by fluorescence in situ hybridization (FISH) was negative in all four cases, excluding well-differentiated liposarcoma. Radiologists should be familiar with PMPF to provide appropriate guidance on clinical management.

Localized Multifocal Retroperitoneal Ganglioneuroma with an Infiltrative Appearance on Imaging: A Case Report.

Multifocal ganglioneuromas are characterized by the presence of multiple benign neuroepithelial tumor nodules and are less common than solitary tumors. A small percentage of ganglioneuromas present with a fatty appearance. Only a few cases of multifocal ganglioneuromas have been reported, due to both their rarity and minimal symptomatic presentation; therefore, generalizations about risk factors and predictive markers are very difficult. Here, we report a case of multifocal retroperitoneal ganglioneuroma with an infiltrative appearance on computed tomography (CT). The tumor demonstrated slow growth on multiple imaging studies and was associated with abdominal and flank pain. The aggressive appearance eventually led to surgical resection 18 months after the initial incidental finding on CT. Postsurgical analysis of the tumor on imaging was crucial in revealing its nodularity and infiltration, as well as for clarifying its retroperitoneal location inseparable from the adrenal gland. Histology demonstrated Schwann cells and ganglion cells without atypia or increased cellularity, and with no mitosis or necrosis seen. Our case highlights the consideration of ganglioneuroma with fatty infiltration in the differential diagnosis of a fatty tumor in the mediastinum or retroperitoneum. Additionally, our report differentiates multifocal ganglioneuroma with fatty infiltration from lipomatous ganglioneuroma on radiology and histopathology.

What the radiologist needs to know about urolithiasis: part 1--pathogenesis, types, assessment, and variant anatomy.

OBJECTIVE: This article reviews types of urinary calculi and their imaging appearances, presents direct and secondary imaging findings of urolithiasis, and provides an overview of treatment methods. Pertinent imaging findings that impact clinical management are highlighted. The implications of complex or variant genitourinary anatomy are reviewed. We outline a standard format for the reporting of urolithiasis to facilitate informed clinical management decisions. CONCLUSION: Unenhanced CT is the preferred examination for evaluation of urolithiasis because of its availability, ease of performance, and high sensitivity. An awareness of the important imaging findings to report allows appropriate and efficient therapy.

What the radiologist needs to know about urolithiasis: part 2--CT findings, reporting, and treatment.

OBJECTIVE: This article reviews types of urinary calculi and their imaging appearances, presents direct and secondary imaging findings of urolithiasis, and provides an overview of treatment methods. Pertinent imaging findings that affect clinical management are highlighted. The implications of complex or variant genitourinary anatomy are reviewed. We outline a standard format for the reporting of urolithiasis to facilitate informed clinical management decisions. CONCLUSION: Unenhanced CT is the preferred examination for evaluation of urolithiasis because of its availability, ease of performance, and high sensitivity. An awareness of the important imaging findings to report allows appropriate and efficient therapy.