OMERACT validation of a deep learning algorithm for automated absolute quantification of knee joint effusion versus manual semi-quantitative assessment.

OBJECTIVE: To begin evaluating deep learning (DL)-automated quantification of knee joint effusion-synovitis via the OMERACT filter. METHODS: A DL algorithm previously trained on Osteoarthritis Initiative (OAI) knee MRI automatically quantified effusion volume in MRI of 53 OAI subjects, which were also scored semi-quantitatively via KIMRISS and MOAKS by 2-6 readers. RESULTS: DL-measured knee effusion correlated significantly with experts' assessments (Kendall's tau 0.34-0.43) CONCLUSION: The close correlation of automated DL knee joint effusion quantification to KIMRISS manual semi-quantitative scoring demonstrated its criterion validity. Further assessments of discrimination and truth vs. clinical outcomes are still needed to fully satisfy OMERACT filter requirements.

Unsupervised Multi-Latent Space RL Framework for Video Summarization in Ultrasound Imaging.

The COVID-19 pandemic has highlighted the need for a tool to speed up triage in ultrasound scans and provide clinicians with fast access to relevant information. To this end, we propose a new unsupervised reinforcement learning (RL) framework with novel rewards to facilitate unsupervised learning by avoiding tedious and impractical manual labelling for summarizing ultrasound videos. The proposed framework is capable of delivering video summaries with classification labels and segmentations of key landmarks which enhances its utility as a triage tool in the emergency department (ED) and for use in telemedicine. Using an attention ensemble of encoders, the high dimensional image is projected into a low dimensional latent space in terms of: a) reduced distance with a normal or abnormal class (classifier encoder), b) following a topology of landmarks (segmentation encoder), and c) the distance or topology agnostic latent representation (autoencoders). The summarization network is implemented using a bi-directional long short term memory (Bi-LSTM) which utilizes the latent space representation from the encoder. Validation is performed on lung ultrasound (LUS), that typically represent potential use cases in telemedicine and ED triage acquired from different medical centers across geographies (India and Spain). The proposed approach trained and tested on 126 LUS videos showed high agreement with the ground truth with an average precision of over 80% and average F1 score of well over 44 ±1.7 %. The approach resulted in an average reduction in storage space of 77% which can ease bandwidth and storage requirements in telemedicine.

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.

Integrating User-Input into Deep Convolutional Neural Networks for Thyroid Nodule Segmentation.

Delineation of thyroid nodule boundaries is necessary for cancer risk assessment and accurate categorization of nodules. Clinicians often use manual or bounding-box approach for nodule assessment which leads to subjective results. Consequently, agreement in thyroid nodule categorization is poor even among experts. Computer-aided diagnosis systems could reduce this variability by minimizing the extent of user interaction and by providing precise nodule segmentations. In this study, we present a novel approach for effective thyroid nodule segmentation and tracking using a single user click on the region of interest. When a user clicks on an ultrasound sweep, our proposed model can predict nodule segmentation over the entire sequence of frames. Quantitative evaluations show that the proposed method out-performs the bounding box approach in terms of the dice score on a large dataset of 372 ultrasound images. The proposed approach saves expert time and reduces the potential variability in thyroid nodule assessment. The proposed one-click approach can save clinicians time required for annotating thyroid nodules within ultrasound images/sweeps. With minimal user interaction we would be able to identify the nodule boundary which can further be used for volumetric measurement and characterization of the nodule. This approach can also be extended for fast labeling of large thyroid imaging datasets suitable for training machine-learning based algorithms.

vid-SAMGRAH: A PyTorch framework for multi-latent space reinforcement learning driven video summarization in ultrasound imaging.

The COVID-19 pandemic has accelerated the need for automatic triaging and summarization of ultrasound videos for fast access to pathologically relevant information in the Emergency Department and lowering resource requirements for telemedicine. In this work, a PyTorch based unsupervised reinforcement learning methodology which incorporates multi feature fusion to output classification labels, segmentation maps and summary videos for lung ultrasound is presented. The use of unsupervised training eliminates tedious manual labeling of key-frames by clinicians opening new frontiers in scalability in training using unlabeled or weakly labeled data. Our approach was benchmarked against expert clinicians from different geographies displaying superior Precision and F1 scores (over 80% and 44%).

Automated thyroid nodule detection from ultrasound imaging using deep convolutional neural networks.

Thyroid cancer is the most common endocrine cancer and its incidence has continuously increased worldwide. In this paper, we focus on the challenging problem of nodule detection from ultrasound scans. In current clinical practice, this task is performed manually, which is tedious, subjective and highly depends on the clinical experience of radiologists. We propose a novel deep neural network architecture with carefully designed loss function regularization, and network hyperparameters to perform nodule detection without complex post-processing refinement steps. The local training and validation datasets consist of 2461 and 820 ultrasound frames acquired from 60 and 20 patients with a high degree of variability, respectively. The core of the proposed method is a deep learning framework based on multi-task model Mask R-CNN. We have developed a loss function with regularization that prioritizes detection over segmentation. Validation was conducted for 821 ultrasound frames from 20 patients. The proposed model can detect various types of thyroid nodules. The experimental results indicate that our proposed method is effective in thyroid nodule detection. Comparisons with the results by Faster R-CNN and conventional Mask R-CNN demonstrate that the proposed model outperforms the prior state-of-the-art detection methods.

Right Ventricular Segmentation from MRI Using Deep Convolutional Neural Networks.

The assessment of right ventricular (RV) function is essential in the diagnosis of many cardiac diseases. Magnetic resonance imaging (MRI) offers an excellent solution to image right ventricle non-invasively with high contrast and temporal resolution. Manual assessment of the RV function from MRI sequences is tedious and time-consuming and automating the process is of great interest. This study proposes a convolutional neural network-based machine learning approach to automate the delineation of the RV from a sequence of MRI. The architecture of the neural network differs from that of a widely-known U-Net approach. Additionally, the proposed approach used image concatenation to create and utilize 3D spatial information in the segmentation process. Quantitative evaluations were performed over 256 images acquired from 16 patients in publicly available data in comparison to manual delineations. Comparisons with the results by U-Net demonstrated that the proposed method outperforms the prior state-of-the-art method.

Multiview Three-Dimensional Echocardiography Image Fusion Using a Passive Measurement Arm.

Three-dimensional (3D) echocardiography offers a fast and efficient way to scan and assess the structures and function of the heart. However, due to limitations inherent to 3D echocardiography such as limited field-of-view and low signal-to-noise ratio, 3D assessment of the heart is performed only in a minority of patients who undergo transthoracic echocardiography. One approach for improving the field-of-view and image quality is to scan the heart from multiple locations by moving the transducer and fusing the resulting images into a single volume, which requires 3D alignment of individual volumetric echocardiography scans. Previous approaches relied on optical or electromagnetic trackers for transducer tracking. This study proposes a passive measurement arm system for tracking the position of the ultrasound transducer and thereby aligning multiple echocardiography scans. The proposed system does not suffer from line-of-sight limitation as in the case of an optical tracking based fusion system. Additionally, in contrast to an electromagnetic based tracking system, measurement arm measurements are not affected by other ferromagnetic materials in the vicinity. The proposed approach was tested by scanning a heart phantom and fusing nine echocardiography volumes acquired from different locations. The fusion of all nine scans yielded a percentage field-of-view improvement of 98.5%.

Hip Joint Contact Pressure Distribution During Pavlik Harness Treatment of an Infant Hip: A Patient-Specific Finite Element Model.

Developmental dysplasia of the hip (DDH) in infants under 6 months of age is typically treated by the Pavlik harness (PH). During successful PH treatment, a subluxed/dislocated hip is spontaneously reduced into the acetabulum, and DDH undergoes self-correction. PH treatment may fail due to avascular necrosis (AVN) of the femoral head. An improved understanding of mechanical factors accounting for the success/failure of PH treatment may arise from investigating articular cartilage contact pressure (CCP) within a hip during treatment. In this study, CCP in a cartilaginous infant hip was investigated through patient-specific finite element (FE) modeling. We simulated CCP of the hip equilibrated at 90 deg flexion at abduction angles of 40 deg, 60 deg, and 80 deg. We found that CCP was predominantly distributed on the anterior and posterior acetabulum, leaving the superior acetabulum (mainly superolateral) unloaded. From a mechanobiological perspective, hypothesizing that excessive pressure inhibits growth, our results qualitatively predicted increased obliquity and deepening of the acetabulum under such CCP distribution. This is the desired and observed therapeutic effect in successful PH treatment. The results also demonstrated increase in CCP as abduction increased. In particular, the simulation predicted large magnitude and concentrated CCP on the posterior wall of the acetabulum and the adjacent lateral femoral head at extreme abduction (80 deg). This CCP on lateral femoral head may reduce blood flow in femoral head vessels and contribute to AVN. Hence, this study provides insight into biomechanical factors potentially responsible for PH treatment success and complications.

Multiview 3-D Echocardiography Fusion with Breath-Hold Position Tracking Using an Optical Tracking System.

Recent advances in echocardiography allow real-time 3-D dynamic image acquisition of the heart. However, one of the major limitations of 3-D echocardiography is the limited field of view, which results in an acquisition insufficient to cover the whole geometry of the heart. This study proposes the novel approach of fusing multiple 3-D echocardiography images using an optical tracking system that incorporates breath-hold position tracking to infer that the heart remains at the same position during different acquisitions. In six healthy male volunteers, 18 pairs of apical/parasternal 3-D ultrasound data sets were acquired during a single breath-hold as well as in subsequent breath-holds. The proposed method yielded a field of view improvement of 35.4 ± 12.5%. To improve the quality of the fused image, a wavelet-based fusion algorithm was developed that computes pixelwise likelihood values for overlapping voxels from multiple image views. The proposed wavelet-based fusion approach yielded significant improvement in contrast (66.46 ± 21.68%), contrast-to-noise ratio (49.92 ± 28.71%), signal-to-noise ratio (57.59 ± 47.85%) and feature count (13.06 ± 7.44%) in comparison to individual views.

Cross-Modality Validation of Acetabular Surface Models Using 3-D Ultrasound Versus Magnetic Resonance Imaging in Normal and Dysplastic Infant Hips.

Current imaging diagnosis of developmental dysplasia of the hip (DDH) in infancy relies on 2-D ultrasound (US), which is highly operator-dependent. 3-D US offers more complete, and potentially more reliable, imaging of infant hip geometry. We sought to validate the fidelity of acetabular surface models obtained by 3-D US against those obtained concurrently by magnetic resonance imaging (MRI). 3-D US and MRI scans were performed on the same d in 20 infants with normal to severely dysplastic hips (mean age, 57 d; range 13-181 d). 3-D US was performed by two observers using a Philips VL13-5 probe. Coronal 3-D multi-echo data image combination (MEDIC) magnetic resonance (MR) images (1-mm slice thickness) were obtained, usually without sedation, in a 1.5 T Siemens unit. Acetabular surface models were generated for 40 hips from 3-D US and MRI using semi-automated tracing software, separately by three observers. For each hip, the 3-D US and MRI models were co-registered to overlap as closely as possible using Amira software, and the root mean square (RMS) distances between points on the models were computed. 3-D US scans took 3.2 s each. Inter-modality variability was visually minimal. Mean RMS distance between corresponding points on the acetabular surface at 3-D US and MRI was 0.4 ± 0.3 mm, with 95% confidence interval <1 mm. Mean RMS errors for inter-observer and intra-observer comparisons were significantly less for 3-D US than for MRI, while inter-scan and inter-modality comparisons showed no significant difference. Acetabular geometry was reproduced by 3-D US surface models within 1 mm of the corresponding 3-D MRI surface model, and the 3-D US models were more reliable. This validates the fidelity of 3-D US modeling and encourages future use of 3-D US in assessing infant acetabulum anatomy, which may be useful to detect and monitor treatment of hip dysplasia.

Multiview echocardiography fusion using an electromagnetic tracking system.

Three-dimensional ultrasound is an emerging modality for the assessment of complex cardiac anatomy and function. The advantages of this modality include lack of ionizing radiation, portability, low cost, and high temporal resolution. Major limitations include limited field-of-view, reliance on frequently limited acoustic windows, and poor signal to noise ratio. This study proposes a novel approach to combine multiple views into a single image using an electromagnetic tracking system in order to improve the field-of-view. The novel method has several advantages: 1) it does not rely on image information for alignment, and therefore, the method does not require image overlap; 2) the alignment accuracy of the proposed approach is not affected by any poor image quality as in the case of image registration based approaches; 3) in contrast to previous optical tracking based system, the proposed approach does not suffer from line-of-sight limitation; and 4) it does not require any initial calibration. In this pilot project, we were able to show that using a heart phantom, our method can fuse multiple echocardiographic images and improve the field-of view. Quantitative evaluations showed that the proposed method yielded a nearly optimal alignment of image data sets in three-dimensional space. The proposed method demonstrates the electromagnetic system can be used for the fusion of multiple echocardiography images with a seamless integration of sensors to the transducer.

Patient movement compensation for 3D echocardiography fusion.

Limited field of view (FOV) is a major problem for 3D real-time echocardiography (3DRTE), which results in an incomplete representation of cardiac anatomy. Various image registration techniques have been proposed to improve the field of view in 3DRTE by fusing multiple image volumes. However, these techniques require significant overlap between the individual volumes and rely on high image resolution and high signal-to-noise ratio. Changes in the heart position due to patient movement during image acquisition can also reduce the quality of image fusion. In this paper, we propose a multi-camera based optical tracking system which 1) eliminates the need for image overlap and 2) compensates for patient movement during acquisition. We compensate for patient movement by continuously tracking the patient position using skin markers and incorporating this information into the fusion process. We fuse volumes acquired during R-R wave peaks based on Electrocardiogram (ECG) data to account for retrospective image acquisition. The fusion technique was validated using a heart phantom (Shelley Medical Imaging Technologies) and on one healthy volunteer. The fused ultrasound volumes could be generated in within 2 seconds and were found to have complete myocardial boundaries alignment upon visual assessment. No stitching artefacts or movement related artefacts were observed in the fused image.