Interobserver Variability of Hip Dysplasia Indices on Sweep Ultrasound for Novices, Experts, and Artificial Intelligence.

BACKGROUND: Ultrasound for developmental dysplasia of the hip (DDH) is challenging for nonexperts to perform and interpret. Recording "sweep" images allows more complete hip assessment, suitable for automation by artificial intelligence (AI), but reliability has not been established. We assessed agreement between readers of varying experience and a commercial AI algorithm, in DDH detection from infant hip ultrasound sweeps. METHODS: We selected a full spectrum of poor-to-excellent quality images and normal to severe dysplasia, in 240 hips (120 single 2-dimensional images, 120 sweeps). For 12 readers (radiologists, sonographers, clinicians and researchers; 3 were DDH subspecialists), and a ultrasound-FDA-cleared AI software package (Medo Hip), we calculated interobserver reliability for alpha angle measurements by intraclass correlation coefficient (ICC2,1) and for DDH classification by Randolph Kappa. RESULTS: Alpha angle reliability was high for AI versus subspecialists (ICC=0.87 for sweeps, 0.90 for single images). For DDH diagnosis from sweeps, agreement was high between subspecialists (kappa=0.72), and moderate for nonsubspecialists (0.54) and AI (0.47). Agreement was higher for single images (kappa=0.80, 0.66, 0.49). AI reliability deteriorated more than human readers for the poorest-quality images. The agreement of radiologists and clinicians with the accepted standard, while still high, was significantly poorer for sweeps than 2D images (P<0.05). CONCLUSIONS: In a challenging exercise representing the wide spectrum of image quality and reader experience seen in real-world hip ultrasound, agreement on DDH diagnosis from easily obtained sweeps was only slightly lower than from single images, likely because of the additional step of selecting the best image. AI performed similarly to a nonsubspecialist human reader but was more affected by low-quality images.

Cardiac MRI-Derived Myocardial Deformation Parameters Correlate with Pulmonary Valve Replacement Indications in Repaired Tetralogy of Fallot.

Right ventricular (RV) volumetric cardiac magnetic resonance (CMR) criteria serve as indicators for pulmonary valve replacement (PVR) in repaired tetralogy of Fallot (rTOF). Myocardial deformation and tricuspid valve displacement parameters may be more sensitive measures of RV dysfunction. This study's aim was to describe rTOF RV deformation and tricuspid displacement patterns using novel CMR semi-automated software and determine associations with standard CMR measures. Retrospective study of 78 pediatric rTOF patients was compared to 44 normal controls. Global RV longitudinal and circumferential strain and strain rate (SR) and tricuspid valve (TV) displacement were measured. Correlation analysis between strain, SR, TV displacement, and volumes was performed between and within subgroups. The sensitivity and specificity of strain parameters in predicting CMR criteria for PVR was determined. Deformation variables were reduced in rTOF compared to controls. Decreased RV strain and TV shortening were associated with increased RV volumes and decreased RVEF. Longitudinal and circumferential parameters were predictive of RVESVi (> 80 ml/m2) and RVEF (< 47%), with circumferential strain (> - 15.88%) and SR (> - 0.62) being most sensitive. Longitudinal strain was unchanged between rTOF subgroups, while circumferential strain trended abnormal in those meeting PVR criteria compared to controls. RV deformation and TV displacement are abnormal in rTOF, and RV circumferential strain variation may reflect an adaptive response to chronic volume or pressure load. This coupled with associations of ventricular deformation with traditional PVR indications suggest importance of this analysis in the evolution of rTOF RV assessment.

Quantification of circumferential, longitudinal, and radial global fractional shortening using steady-state free precession cines: a comparison with tissue-tracking strain and application in Fabry disease.

PURPOSE: Conventional calculation of myocardial strain requires tissue-tracking. A surrogate for strain called global fractional shortening (GFS) is proposed based on changes in dimensions of endocardial and epicardial surfaces without tissue-tracking. METHODS: Three-dimensional endocardial and epicardial left ventricular surfaces traced at end-diastole and end-systole using conventional steady-state free precession cine images were used to calculate GFScc (circumferential), GFSll (longitudinal), and GFSrr (radial) using fractional length changes in each direction over the heart surface. GFS values were validated using finite element models (FEM) and in vivo using tagging-derived strains (εcc ,εll ,εrr ) in patients with a wide range of ejection fraction (EF) and diagnosis (n=32). GFS was also measured in 31 patients with Fabry disease and matched healthy controls. RESULTS: GFS values were within 3% of average FEM-derived Lagrangian strains and had good agreement in vivo (GFScc =-14 ± 4%, εcc =-14 ± 4%, R(2) =0.85; GFSll =-12 ± 4%, εll =-12 ± 4%, R(2) =0.72; GFSrr =46 ± 21%). εrr could not be measured reliably from tagging. Compared with healthy controls with matched EF, patients with Fabry disease had significantly increased GFScc (Endo) (-28 ± 3% versus -25 ± 2%), decreased GFScc(Epi) (-10 ± 2% versus -11 ± 2%) and decreased GFSll for all components. CONCLUSION: GFS yields similar values to conventionally measured strains without requiring tissue-tracking. Compared with controls, patients with Fabry disease have significant differences in several GFS components.

Ultrasonic mapping of midpalatal suture - An ex-vivo study.

OBJECTIVE: Rapid maxillary expansion is a common orthodontic procedure to correct maxillary constriction. Assessing the midpalatal suture (MPS) expansion plays a crucial role in treatment planning to determine its effectiveness. The objectives of this preliminary investigation are to demonstrate a proof of concept that the palatal bone underlying the rugae can be clearly imaged by ultrasound (US) and the reconstructed axial view of the US image accurately maps the MPS patency. METHODS: An ex-vivo US scanning was conducted on the upper jawbones of two piglet's carcasses before and after the creation of bone defects, which simulated the suture opening. The planar images were processed to enhance bone intensity distribution before being orderly stacked to fuse into a volume. Graph-cut segmentation was applied to delineate the palatal bone to generate a bone volume. The accuracy of the reconstructed bone volume and the suture opening was validated by the micro-computed tomography (µCT) data used as the ground truth and compared with cone beam computed tomography (CBCT) data as the clinical standard. Also included in the comparison is the rugae thickness. Correlation and Bland-Altman plots were used to test the agreement between the two methods: US versus µCT/CBCT. RESULTS: The reconstruction of the US palatal bone volumes was accurate based on surface topography comparison with a mean error of 0.19 mm for pre-defect and 0.15 mm and 0.09 mm for post-defect models of the two samples, respectively when compared with µCT volumes. A strong correlation (R2 ≥ 0.99) in measuring MPS expansion was found between US and µCT/CBCT with MADs of less than 0.05 mm, 0.11 mm and 0.23 mm for US, µCT and CBCT, respectively. CONCLUSIONS: It was possible to axially image the MPS opening and rugae thickness accurately using high-frequency ultrasound. CLINICAL SIGNIFICANCE: This study introduces an ionizing radiation-free, low-cost, and portable technique to accurately image a difficult part of oral cavity anatomy. The advantages of conceivable visualization could promise a successful clinical examination of MPS to support the predictable treatment outcome of maxillary transverse deficiency.

Benefit of stereoscopic volume rendering for the identification of pediatric pulmonary vein stenosis from CT angiography.

The use of three-dimensional (3D) technologies in medical practice is increasing; however, its use is largely untested. One 3D technology, stereoscopic volume-rendered 3D display, can improve depth perception. Pulmonary vein stenosis (PVS) is a rare cardiovascular pathology, often diagnosed by computed tomography (CT), where volume rendering may be useful. Depth cues may be lost when volume rendered CT is displayed on regular screens instead of 3D displays. The objective of this study was to determine whether the 3D stereoscopic display of volume-rendered CT improved perception compared to standard monoscopic display, as measured by PVS diagnosis. CT angiograms (CTAs) from 18 pediatric patients aged 3 weeks to 2 years were volume rendered and displayed with and without stereoscopic display. Patients had 0 to 4 pulmonary vein stenoses. Participants viewed the CTAs in 2 groups with half on monoscopic and half on stereoscopic display and the converse a minimum of 2 weeks later, and their diagnoses were recorded. A total of 24 study participants, comprised of experienced staff cardiologists, cardiovascular surgeons and radiologists, and their trainees viewed the CTAs and assessed the presence and location of PVS. Cases were classified as simple (2 or fewer lesions) or complex (3 or more lesions). Overall, there were fewer type 2 errors in diagnosis for stereoscopic display than standard display, an insignificant difference (p = 0.095). There was a significant decrease in type 2 errors for complex multiple lesion cases (≥3) vs simpler cases (p = 0.027) and improvement in localization of pulmonary veins (p = 0.011). Subjectively, 70% of participants stated that stereoscopy was helpful in the identification of PVS. The stereoscopic display did not result in significantly decreased errors in PVS diagnosis but was helpful for more complex cases.

Deep Learning Based Parameterization of Diffeomorphic Image Registration for Cardiac Image Segmentation.

Cardiac segmentation from magnetic resonance imaging (MRI) is one of the essential tasks in analyzing the anatomy and function of the heart for the assessment and diagnosis of cardiac diseases. However, cardiac MRI generates hundreds of images per scan, and manual annotation of them is challenging and time-consuming, and therefore processing these images automatically is of interest. This study proposes a novel end-to-end supervised cardiac MRI segmentation framework based on a diffeomorphic deformable registration that can segment cardiac chambers from 2D and 3D images or volumes. To represent actual cardiac deformation, the method parameterizes the transformation using radial and rotational components computed via deep learning, with a set of paired images and segmentation masks used for training. The formulation guarantees transformations that are invertible and prevents mesh folding, which is essential for preserving the topology of the segmentation results. A physically plausible transformation is achieved by employing diffeomorphism in computing the transformations and activation functions that constrain the range of the radial and rotational components. The method was evaluated over three different data sets and showed significant improvements compared to exacting learning and non-learning based methods in terms of the Dice score and Hausdorff distance metrics.

Pediatric patient-specific three-dimensional virtual models for surgical decision making in resection of hepatic and retroperitoneal tumors.

PURPOSE: Typically, preoperative imaging is viewed in two dimensions (2D) only, but three-dimensional (3D) virtual models may improve viewers' anatomical perspective by permitting them to interact with the imaging through manipulating it in space. Research into the utility of these models in most surgical specialties is growing rapidly. This study investigates the utility of 3D virtual models of complex pediatric abdominal tumors for clinical decision making, particularly the decision to proceed with surgical resection or not. METHODS: 3D virtual models of tumors and adjacent anatomy were created from CT images of pediatric patients scanned for Wilms tumor, neuroblastoma or hepatoblastoma. Pediatric surgeons individually assessed the resectability of the tumors. First, they assessed resectability using the standard protocol of viewing imaging on conventional screens and then reassessed resectability after being presented with the 3D virtual models. Inter-physician agreement on resectability for each patient was analyzed using Krippendorff's alpha. Inter-physician agreement was used as a surrogate for correct interpretation. Participants were also surveyed afterward on the utility and practicality of the 3D virtual models for clinical decision making. RESULTS: Inter-physician agreement when using CT imaging alone was "fair" (Krippendorff's alpha α = 0.399), while inter-physician agreement when using 3D virtual models increased to "moderate" (Krippendorff's alpha α = 0.532). When surveyed about model utility, all 5 participants considered them helpful. Two participants felt the models would be practical for clinical use in most cases, while 3 felt they would be practical for select cases only. CONCLUSION: This study demonstrates the subjective utility of 3D virtual models of pediatric abdominal tumors for clinical decision making. The models are an adjunct that can be particularly useful in complicated tumors that efface or displace critical structures that may impact resectability. Statistical analysis demonstrates the improved inter-rater agreement with the 3D stereoscopic display over the 2D display. The use of 3D displays of medical images will increase over time, and evaluation of their potential usefulness in various clinical settings is necessary.

ACU2E-Net: A novel predict-refine attention network for segmentation of soft-tissue structures in ultrasound images.

Segmentation of anatomical structures in ultrasound images is a challenging task due to existence of artifacts inherit to the modality such as speckle noise, attenuation, shadowing, uneven textures and blurred boundaries. This paper presents a novel attention-based predict-refine network, called ACU2E-Net, for segmentation of soft-tissue structures in ultrasound images. The network consists of two modules: a predict module, which is built upon our newly proposed attentive coordinate convolution; and a novel multi-head residual refinement module, which consists of three parallel residual refinement modules. The attentive coordinate convolution is designed to improve the segmentation accuracy by perceiving the shape and positional information of the target anatomy. The proposed multi-head residual refinement module reduces both segmentation biases and variances by integrating residual refinement and ensemble strategies. Moreover, it avoids multi-pass training and inference commonly seen in ensemble methods. To show the effectiveness of our method, we collect a comprehensive dataset of thyroid ultrasound scans from 12 different imaging centers, and evaluate our proposed network against state-of-the-art segmentation methods. Comparisons against state-of-the-art models demonstrate the competitive performance of our newly designed network on both the transverse and sagittal thyroid images. Ablation studies show that proposed modules improve the segmentation Dice score of the baseline model from 79.62% to 80.97% and 82.92% while reducing the variance from 6.12% to 4.67% and 3.21% in transverse and sagittal views, respectively.

MyoPS: A benchmark of myocardial pathology segmentation combining three-sequence cardiac magnetic resonance images.

Assessment of myocardial viability is essential in diagnosis and treatment management of patients suffering from myocardial infarction, and classification of pathology on the myocardium is the key to this assessment. This work defines a new task of medical image analysis, i.e., to perform myocardial pathology segmentation (MyoPS) combining three-sequence cardiac magnetic resonance (CMR) images, which was first proposed in the MyoPS challenge, in conjunction with MICCAI 2020. Note that MyoPS refers to both myocardial pathology segmentation and the challenge in this paper. The challenge provided 45 paired and pre-aligned CMR images, allowing algorithms to combine the complementary information from the three CMR sequences for pathology segmentation. In this article, we provide details of the challenge, survey the works from fifteen participants and interpret their methods according to five aspects, i.e., preprocessing, data augmentation, learning strategy, model architecture and post-processing. In addition, we analyze the results with respect to different factors, in order to examine the key obstacles and explore the potential of solutions, as well as to provide a benchmark for future research. The average Dice scores of submitted algorithms were 0.614±0.231 and 0.644±0.153 for myocardial scars and edema, respectively. We conclude that while promising results have been reported, the research is still in the early stage, and more in-depth exploration is needed before a successful application to the clinics. MyoPS data and evaluation tool continue to be publicly available upon registration via its homepage (www.sdspeople.fudan.edu.cn/zhuangxiahai/0/myops20/).

Deep Learning Segmentation of the Right Ventricle in Cardiac MRI: The M&Ms Challenge.

In recent years, several deep learning models have been proposed to accurately quantify and diagnose cardiac pathologies. These automated tools heavily rely on the accurate segmentation of cardiac structures in MRI images. However, segmentation of the right ventricle is challenging due to its highly complex shape and ill-defined borders. Hence, there is a need for new methods to handle such structure's geometrical and textural complexities, notably in the presence of pathologies such as Dilated Right Ventricle, Tricuspid Regurgitation, Arrhythmogenesis, Tetralogy of Fallot, and Inter-atrial Communication. The last MICCAI challenge on right ventricle segmentation was held in 2012 and included only 48 cases from a single clinical center. As part of the 12th Workshop on Statistical Atlases and Computational Models of the Heart (STACOM 2021), the M&Ms-2 challenge was organized to promote the interest of the research community around right ventricle segmentation in multi-disease, multi-view, and multi-center cardiac MRI. Three hundred sixty CMR cases, including short-axis and long-axis 4-chamber views, were collected from three Spanish hospitals using nine different scanners from three different vendors, and included a diverse set of right and left ventricle pathologies. The solutions provided by the participants show that nnU-Net achieved the best results overall. However, multi-view approaches were able to capture additional information, highlighting the need to integrate multiple cardiac diseases, views, scanners, and acquisition protocols to produce reliable automatic cardiac segmentation algorithms.

Deep Learning using Pre-Brachytherapy MRI to Automatically Predict Applicator Induced Complex Uterine Deformation.

This novel deep-learning (DL) algorithm addresses the challenging task of predicting uterine shape and location when deformed from its natural anatomy by the presence of an intrauterine (tandem)/intravaginal (ring) applicator during brachytherapy (BT) treatment for locally advanced cervical cancer. Paired pelvic MRI datasets from 92 subjects, acquired without (pre-BT) and with (at-BT) applicators, were used. We propose a novel automated algorithm to segment the uterus in pre-BT MR images using a deep convolutional neural network (CNN) incorporated with autoencoders. The proposed neural net is based on a pre-trained CNN Inception V4 architecture. It predicts a compressed vector by applying a multi-layer autoencoder, which is then back-projected into the segmentation contour of the uterus. Following this, another transfer learning approach using a modified U-net model is employed to predict the at-BT uterus shape from pre-BT MRI. The complex and large deformations of the uterus are quantified using free form deformation method. The proposed algorithm yielded an average Dice Coefficient (DC) of 94.1±3.3 and an average Hausdorff Distance (HD) of 4.0±3.1 mm compared to the manually defined ground truth by expert clinicians. Further, the modified U-net prediction of the at-BT uterus resulted in a DC accuracy of 88.1±3.8 and HD of 5.8±3.6 mm. The mean uterine surface point-to-point displacement was 25.0 [10.0-62.5] mm from the pre-BT position. Our unique DL method can thus successfully predict tandem-deformed uterine shape and position from MR images taken before the BT implant procedure i.e. without the applicator in place. Clinical relevance-The proposed DL-based framework can be incorporated as an automatic prediction tool of uterine deformation due to applicator insertion for personalized BT treatments. It holds promise for more streamlined clinical/technical decision-making before BT applicator insertion resulting in improved dosimetric outcomes.

Normal variation of infant hip development : patterns revealed by 3D ultrasound.

AIMS: Studies of infant hip development to date have been limited by considering only the changes in appearance of a single ultrasound slice (Graf's standard plane). We used 3D ultrasound (3DUS) to establish maturation curves of normal infant hip development, quantifying variation by age, sex, side, and anteroposterior location in the hip. METHODS: We analyzed 3DUS scans of 519 infants (mean age 64 days (6 to 111 days)) presenting at a tertiary children's hospital for suspicion of developmental dysplasia of the hip (DDH). Hips that did not require ultrasound follow-up or treatment were classified as 'typically developing'. We calculated traditional DDH indices like α angle (αSP), femoral head coverage (FHCSP), and several novel indices from 3DUS like the acetabular contact angle (ACA) and osculating circle radius (OCR) using custom software. RESULTS: α angle, FHC, and ACA indices increased and OCR decreased significantly by age in the first four months, mean αSP rose from 62.2° (SD 5.7°) to 67.3° (SD 5.2°) (p < 0.001) in one- to eight- and nine- to 16-week-old infants, respectively. Mean αSP and mean FHCSP were significantly, but only slightly, lower in females than in males. There was no statistically significant difference in DDH indices observed between left and right hip. All 3DUS indices varied significantly between anterior and posterior section of the hip. Mean 3D indices of α angle and FHC were significantly lower anteriorly than posteriorly: αAnt = 58.2° (SD 6.1°), αPost = 63.8° (SD 6.3°) (p < 0.001), FHCAnt = 43.0 (SD 7.4), and FHCPost = 55.4° (SD 11.2°) (p < 0.001). Acetabular rounding measured byOCR indices was significantly greater in the anterior section of the hip (p < 0.001). CONCLUSION: We used 3DUS to show that hip shape and normal growth pattern vary significantly between anterior and posterior regions, by magnitudes similar to age-related changes. This highlights the need for careful selection of the Graf plane during 2D ultrasound examination. Whole-joint evaluation by obtaining either 3DUS or manual 'sweep' video images provides more comprehensive DDH assessment.Cite this article: Bone Jt Open 2022;3(11):913-923.

A New Semi-automated Algorithm for Volumetric Segmentation of the Left Ventricle in Temporal 3D Echocardiography Sequences.

PURPOSE: Echocardiography is commonly used as a non-invasive imaging tool in clinical practice for the assessment of cardiac function. However, delineation of the left ventricle is challenging due to the inherent properties of ultrasound imaging, such as the presence of speckle noise and the low signal-to-noise ratio. METHODS: We propose a semi-automated segmentation algorithm for the delineation of the left ventricle in temporal 3D echocardiography sequences. The method requires minimal user interaction and relies on a diffeomorphic registration approach. Advantages of the method include no dependence on prior geometrical information, training data, or registration from an atlas. RESULTS: The method was evaluated using three-dimensional ultrasound scan sequences from 18 patients from the Mazankowski Alberta Heart Institute, Edmonton, Canada, and compared to manual delineations provided by an expert cardiologist and four other registration algorithms. The segmentation approach yielded the following results over the cardiac cycle: a mean absolute difference of 1.01 (0.21) mm, a Hausdorff distance of 4.41 (1.43) mm, and a Dice overlap score of 0.93 (0.02). CONCLUSION: The method performed well compared to the four other registration algorithms.

Temporomandibular joint segmentation in MRI images using deep learning.

OBJECTIVES: Temporomandibular joint (TMJ) internal derangements (ID) represent the most prevalent temporomandibular joint disorder (TMD) in the population and its diagnosis typically relies on magnetic resonance imaging (MRI). TMJ articular discs in MRIs usually suffer from low resolution and contrast, and it is difficult to identify them. In this study, we applied two convolutional neural networks (CNN) to delineate mandibular condyle, articular eminence, and TMJ disc in MRI images. METHODS: The models were trained on MRI images from 100 patients and validated on images from 40 patients using 2D slices and 3D volume as input, respectively. Data augmentation and five-fold cross-validation scheme were applied to further regularize the models. The accuracy of the models was then compared with four raters having different expertise in reading TMJ-MRI images to evaluate the performance of the models. RESULTS: Both models performed well in segmenting the three anatomical structures. A Dice coefficient of about 0.7 for the articular disc, more than 0.9 for the mandibular condyle, and Hausdorff distance of about 2mm for the articular eminence were achieved in both models. The models reached near-expert performance for the segmentation of TMJ articular disc and performed close to the expert in the segmentation of mandibular condyle and articular eminence. They also surpassed non-experts in segmenting the three anatomical structures. CONCLUSION: This study demonstrated that CNN-based segmentation models can be a reliable tool to assist clinicians identifying key anatomy on TMJ-MRIs. The approach also paves the way for automatic diagnosis of TMD. CLINICAL SIGNIFICANCE: Accurately locating the articular disc is the hardest and most crucial step in the interpretation of TMJ-MRIs and consequently in the diagnosis of TMJ-ID. Automated software that assists in locating the articular disc and its surrounding structures would improve the reliability of TMJ-MRI interpretation, save time and assist in reader training. It will also serve as a foundation for additional automated analysis of pathology in TMJ structures to aid in TMD diagnosis.

Fully automated left atrium segmentation from anatomical cine long-axis MRI sequences using deep convolutional neural network with unscented Kalman filter.

This study proposes a fully automated approach for the left atrial segmentation from routine cine long-axis cardiac magnetic resonance image sequences using deep convolutional neural networks and Bayesian filtering. The proposed approach consists of a classification network that automatically detects the type of long-axis sequence and three different convolutional neural network models followed by unscented Kalman filtering (UKF) that delineates the left atrium. Instead of training and predicting all long-axis sequence types together, the proposed approach first identifies the image sequence type as to 2, 3 and 4 chamber views, and then performs prediction based on neural nets trained for that particular sequence type. The datasets were acquired retrospectively and ground truth manual segmentation was provided by an expert radiologist. In addition to neural net based classification and segmentation, another neural net is trained and utilized to select image sequences for further processing using UKF to impose temporal consistency over cardiac cycle. A cyclic dynamic model with time-varying angular frequency is introduced in UKF to characterize the variations in cardiac motion during image scanning. The proposed approach was trained and evaluated separately with varying amount of training data with images acquired from 20, 40, 60 and 80 patients. Evaluations over 1515 images with equal number of images from each chamber group acquired from an additional 20 patients demonstrated that the proposed model outperformed state-of-the-art and yielded a mean Dice coefficient value of 94.1%, 93.7% and 90.1% for 2, 3 and 4-chamber sequences, respectively, when trained with datasets from 80 patients.

Automatic spinal cord segmentation from axial-view MRI slices using CNN with grayscale regularized active contour propagation.

Accurate positioning of the responsible segment for patients with cervical spondylotic myelopathy (CSM) is clinically important not only to the surgery but also to reduce the incidence of surgical trauma and complications. Spinal cord segmentation is a crucial step in the positioning procedure. This study proposed a fully automated approach for spinal cord segmentation from 2D axial-view MRI slices of patients with CSM. The proposed method was trained and tested using clinical data from 20 CSM patients (359 images) acquired by the Peking University Third Hospital, with ground truth labeled by professional radiologists. The accuracy of the proposed method was evaluated using quantitative measures, the reliability metric as well as visual assessment. The proposed method yielded a Dice coefficient of 87.0%, Hausdorff distance of 9.7 mm, root-mean-square error of 5.9 mm. Higher conformance with ground truth was observed for the proposed method in comparison to the state-of-the-art algorithms. The results are also statistically significant with p-values calculated between state-of-the-art methods and the proposed methods.

Multi-View 3-D Fusion Echocardiography: Enhancing Clinical Feasibility with a Novel Processing Technique.

A novel system for fusing 3-D echocardiography data sets from complementary acoustic windows was evaluated in 12 healthy volunteers and 12 patients with heart failure. We hypothesized that 3-D fusion would enable 3-D echocardiography in patients with limited acoustic windows. At least nine 3-D data sets were recorded, while three infrared cameras tracked the position and orientation of the transducer and chest respiratory movements. Corresponding 2-D planes of the fused 3-D data sets and of single-view 3-D data sets were assessed for image quality and compared with measurements of left ventricular function obtained with contrast 2-D echocardiography. The signal-to-noise ratio in accurately fused 3-D echocardiography recordings improved by 55% in systole (p < 0.001) and 47% in diastole (p < 0.00001) compared with the apical single-view recordings. The 3-D data sets acquired during short breath holds were successfully fused in 11 of 12 patients. The improvement in endocardial border definition (from 11.7 ± 6.0 to 24.0 ± 3.3, p < 0.01) enabled quantitative assessment of left ventricular function in 10 patients, with no significant difference in ejection fraction compared with contrast 2-D echocardiography. In patients with heart failure and limited acoustic windows, the novel fusion protocol provides 3-D data sets suitable for quantitative analysis of left ventricular function.

Automated detection of pneumonia in lung ultrasound using deep video classification for COVID-19.

There is a crucial need for quick testing and diagnosis of patients during the COVID-19 pandemic. Lung ultrasound is an imaging modality that is cost-effective, widely accessible, and can be used to diagnose acute respiratory distress syndrome in patients with COVID-19. It can be used to find important characteristics in the images, including A-lines, B-lines, consolidation, and pleural effusion, which all inform the clinician in monitoring and diagnosing the disease. With the use of portable ultrasound transducers, lung ultrasound images can be easily acquired, however, the images are often of poor quality. They often require an expert clinician interpretation, which may be time-consuming and is highly subjective. We propose a method for fast and reliable interpretation of lung ultrasound images by use of deep learning, based on the Kinetics-I3D network. Our learned model can classify an entire lung ultrasound scan obtained at point-of-care, without requiring the use of preprocessing or a frame-by-frame analysis. We compare our video classifier against ground truth classification annotations provided by a set of expert radiologists and clinicians, which include A-lines, B-lines, consolidation, and pleural effusion. Our classification method achieves an accuracy of 90% and an average precision score of 95% with the use of 5-fold cross-validation. The results indicate the potential use of automated analysis of portable lung ultrasound images to assist clinicians in screening and diagnosing patients.

Localization of cementoenamel junction in intraoral ultrasonographs with machine learning.

OBJECTIVE: Our goal was to automatically identify the cementoenamel junction (CEJ) location in ultrasound images using deep convolution neural networks (CNNs). METHODS: Three CNNs were evaluated using 1400 images and data augmentation. The training and validation were performed by an experienced nonclinical rater with 1000 and 200 images, respectively. Four clinical raters with different levels of experience with ultrasound tested the networks using the other 200 images. In addition to the comparison of the best approach with each rater, we also employed the simultaneous truth and performance level estimation (STAPLE) algorithm to estimate a ground truth based on all labelings by four clinical raters. The final CEJ location estimate was obtained by taking the first moment of the posterior probability computed using the STAPLE algorithm. The study also computed the machine learning-measured CEJ-alveolar bone crest distance. RESULTS: Quantitative evaluations of the 200 images showed that the comparison of the best approach with the STAPLE-estimate yielded a mean difference (MD) of 0.26 mm, which is close to the comparison with the most experienced nonclinical rater (MD=0.25 mm) but far better than the comparison with clinical raters (MD=0.27-0.33 mm). The machine learning-measured CEJ-alveolar bone crest distances correlated strongly (R = 0.933, p < 0.001) with the manual clinical labeling and the measurements were in good agreement with the 95% Bland-Altman's lines of agreement between -0.68 and 0.57 mm. CONCLUSIONS: The study demonstrated the feasible use of machine learning methodology to localize CEJ in ultrasound images with clinically acceptable accuracy and reliability. Likelihood-weighted ground truth by combining multiple labels by the clinical experts compared favorably with the predictions by the best deep CNN approach. CLINICAL SIGNIFICANCE: Identification of CEJ and its distance from the alveolar bone crest play an important role in the evaluation of periodontal status. Machine learning algorithms can learn from complex features in ultrasound images and have potential to provide a reliable and accurate identification in subsecond. This will greatly assist dental practitioners to provide better point-of-care to patients and enhance the throughput of dental care.

Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation: The M&Ms Challenge.

The emergence of deep learning has considerably advanced the state-of-the-art in cardiac magnetic resonance (CMR) segmentation. Many techniques have been proposed over the last few years, bringing the accuracy of automated segmentation close to human performance. However, these models have been all too often trained and validated using cardiac imaging samples from single clinical centres or homogeneous imaging protocols. This has prevented the development and validation of models that are generalizable across different clinical centres, imaging conditions or scanner vendors. To promote further research and scientific benchmarking in the field of generalizable deep learning for cardiac segmentation, this paper presents the results of the Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation (M&Ms) Challenge, which was recently organized as part of the MICCAI 2020 Conference. A total of 14 teams submitted different solutions to the problem, combining various baseline models, data augmentation strategies, and domain adaptation techniques. The obtained results indicate the importance of intensity-driven data augmentation, as well as the need for further research to improve generalizability towards unseen scanner vendors or new imaging protocols. Furthermore, we present a new resource of 375 heterogeneous CMR datasets acquired by using four different scanner vendors in six hospitals and three different countries (Spain, Canada and Germany), which we provide as open-access for the community to enable future research in the field.