US Images Research Articles

Automated grading of prenatal hydronephrosis severity from segmented kidney ultrasounds using deep learning

Background and motivationsAntenatal or prenatal hydronephrosis (AHN) is a common kidney complication in unborn children. While AHN is generally benign and resolves over time, this condition can inflict serious kidney damage or even organ failure due to excessive waste accumulation in severe cases. Regardless of severity, AHN must be clinically monitored for resolution, with treatment plans and medications being revised according to updated prognoses. Kidney ultrasound images are one of the most common methods of monitoring AHN, but grading of this condition is highly subjective and clinicians may select inappropriate therapies or surgical interventions as a result. New approaches are required to differentiate subjects who can be managed without surgical intervention from those who require life-saving operations. MethodsAn end-to-end deep machine learning framework was developed to sequentially detect ultrasound regions of interest, segment kidneys from US images, and classify AHN severity. We propose the novel Kidney Ultrasound Segmentation Network (KUSNet) for kidney segmentation from ultrasound images, and the Prenatal Hydronephrosis Classification Network (PHCNet) for hydronephrosis severity stratification according to the Society of Fetal Urology (SFU) standards. The ground truth kidney masks were generated by two radiologists with more than five years of working experience while the SFU-based annotations for the AHN severity were done by two senior radiologists and three senior urologists with more than ten years of domain expertise. At each stage, the performance of the proposed models was assessed both quantitatively and qualitatively against state-of-the-art networks in the respective fields. ResultsThe proposed KUSNet for ultrasound kidney segmentation achieved 97.6% accuracy, 97.4% precision, 97.6% recall or sensitivity, 97.5% F1-score, 95.5% IoU or Jaccard score, and 92.1% Dice score, beating several state-of-the-arts. On the other hand, the novel PHCNet reached 93.9% accuracy, 93.7% precision, 93.9% recall, 93.8% specificity, and 89.0% F1-score subject-wise when performing multiclass stratification of AHN severity on segmented kidney regions. ConclusionArtificial intelligence-based tools can reliably classify AHN severity to reduce inter- and intra-observer bias, thereby aiding clinicians in the rapid selection of appropriate treatments and surgeries. Moreover, segmenting kidney regions beforehand significantly boosts AHN severity classification performance.

Intraoral ultrasonography image registration for evaluation of partial edentulous ridge: A methodology and validation study

ObjectivesUltrasound (US) reveals details for diagnosing soft- and hard-tissue dimensions around teeth, implants, and the edentulous ridge, not seen in 2D radiographs. Co-registering free-hand US scans with other 3D modalities presents reliability challenges. This study first aims to develop and validate a registration method to longitudinally reproduce US images of the jawbone on a simulator. In addition, it also evaluates the degree of the anatomical match in humans between US images acquired by the proposed registration method and the commonly used freehand acquisitions in comparison to cone beam computed tomography (CBCT) and intra-oral optical scan (IOS), used as references. MethodsA previously introduced ultrasound phantom was employed as a CBCT-US hybrid, suitable for training and technique development of US guides in edentulous ridges. After establishing feasibility in the phantom, the methodology was validated in a cohort of 24 human subjects (26 cases). Soft tissues were delineated on US and IOS, and hard tissues on US and CBCT. US accuracy and repeatability from both guided and freehand scans (non-guided) was assessed as the average distance between US and the references. ResultsGuided US images resembled the references more closely than freehand (non-guided) scans. Notably, delineation of soft and hard tissues was significantly more accurate when employing guides. In the phantom, guided scans exhibited an absolute mean deviation of 81.8 µm for gingiva and 90.4 µm for bone, whereas non-guided scans showed deviations of 150.4 µm and 177.2 µm, respectively. Similarly, in vivo, guided US outperformed non-guided US, with gingiva deviations of 125 µm and 196 µm, and bone deviations of 354 µm and 554 µm, respectively. ConclusionsBy using a registration method, guided US scans improved repeatability and accuracy of mapping hard and soft tissue of the edentulous ridge when compared to non-guided scans. Clinical relevanceThis guided US imaging method could lay the foundation for longitudinal evaluation of tissue behavior and dimensional changes with improved accuracy.

Simulating federated learning for steatosis detection using ultrasound images

We aimed to implement four data partitioning strategies evaluated with four federated learning (FL) algorithms and investigate the impact of data distribution on FL model performance in detecting steatosis using B-mode US images. A private dataset (153 patients; 1530 images) and a public dataset (55 patient; 550 images) were included in this retrospective study. The datasets contained patients with metabolic dysfunction-associated fatty liver disease (MAFLD) with biopsy-proven steatosis grades and control individuals without steatosis. We employed four data partitioning strategies to simulate FL scenarios and we assessed four FL algorithms. We investigated the impact of class imbalance and the mismatch between the global and local data distributions on the learning outcome. Classification performance was assessed with area under the receiver operating characteristic curve (AUC) on a separate test set. AUCs were 0.93 (95% CI 0.92, 0.94) for source-based partitioning scenario with FedAvg, 0.90 (95% CI 0.89, 0.91) for a centralized model, and 0.83 (95% CI 0.81, 0.85) for a model trained in a single-center scenario. When data was perfectly balanced on the global level and each site had an identical data distribution, the model yielded an AUC of 0.90 (95% CI 0.88, 0.92). When each site contained data exclusively from one single class, irrespective of the global data distribution, the AUC fell in the range of 0.34–0.70. FL applied to B-mode US images provide performance comparable to a centralized model and higher than single-center scenario. Global data imbalance and local data heterogeneity influenced the learning outcome.

Sonography in the diagnosis of peri-implant bone defects: An invitro study on native human mandibles.

The aim of this study on native human cadavers was to compare clinical, sonographic, and radiological measurements of fenestrations, dehiscences, and 3-wall bone defects on implants. The examination was carried out on five human mandibles. After the insertion of 27 implants, dehiscences (n = 14), fenestrations (n = 7) and 3-wall bone defects (n = 6) were prepared in a standardized manner. The direct measurement of the bone defects was carried out with a periodontal probe and the radiological examination was carried out using digital volume tomography (DVT). The ultrasound examination (US) was performed using a clinical 24-MHz US imaging probe. Means and standard deviations of the direct, US, and DVT measurements were calculated. Measurements were statistically compared using the Pearson correlation coefficient and Bland-Altman analysis. Bone defects were on average 3.22 ± 1.58 mm per direct measurement, 2.90 ± 1.47 mm using US, and 2.99 ± 1.52 mm per DVT assessment. Pairwise correlations of these measurements were R = .94 (p < .0001) between direct and US, R = .95 (p < .0001) between DVT and US, and R = .96 (p < .0001) between direct and DVT. The mean differences of the measurements (and 95% CI) between direct and US was 0.41 (-0.47 to 1.29), US and DVT 0.33 (-0.30 to 0.97), and direct and DVT 0.28 (-0.50 to 1.07). All peri-implant bone defects could be identified and sonographically measured. US measurements showed a strong correlation with direct and DVT measurements. The sonographic measurement accuracy was highest for dehiscences, followed by fenestrations and 3-wall bone defects.

Detection and diagnosis of automated breast ultrasound in patients with BI-RADS category 4 microcalcifications: a retrospective study

BackgroundAutomated Breast Ultrasound (AB US) has shown good application value and prospects in breast disease screening and diagnosis. The aim of the study was to explore the ability of AB US to detect and diagnose mammographically Breast Imaging Reporting and Data System (BI-RADS) category 4 microcalcifications.Methods575 pathologically confirmed mammographically BI-RADS category 4 microcalcifications from January 2017 to June 2021 were included. All patients also completed AB US examinations. Based on the final pathological results, analyzed and summarized the AB US image features, and compared the evaluation results with mammography, to explore the detection and diagnostic ability of AB US for these suspicious microcalcifications.Results250 were finally confirmed as malignant and 325 were benign. Mammographic findings including microcalcifications morphology (61/80 with amorphous, coarse heterogeneous and fine pleomorphic, 13/14 with fine-linear or branching), calcification distribution (189/346 with grouped, 40/67 with linear and segmental), associated features (70/96 with asymmetric shadow), higher BI-RADS category with 4B (88/120) and 4 C (73/38) showed higher incidence in malignant lesions, and were the independent factors associated with malignant microcalcifications. 477 (477/575, 83.0%) microcalcifications were detected by AB US, including 223 malignant and 254 benign, with a significantly higher detection rate for malignant lesions (x2 = 12.20, P < 0.001). Logistic regression analysis showed microcalcifications with architectural distortion (odds ratio [OR] = 0.30, P = 0.014), with amorphous, coarse heterogeneous and fine pleomorphic morphology (OR = 3.15, P = 0.037), grouped (OR = 1.90, P = 0.017), liner and segmental distribution (OR = 8.93, P = 0.004) were the independent factors which could affect the detectability of AB US for microcalcifications. In AB US, malignant calcification was more frequent in a mass (104/154) or intraductal (20/32), and with ductal changes (30/41) or architectural distortion (58/68), especially with the both (12/12). BI-RADS category results also showed that AB US had higher sensitivity to malignant calcification than mammography (64.8% vs. 46.8%).ConclusionsAB US has good detectability for mammographically BI-RADS category 4 microcalcifications, especially for malignant lesions. Malignant calcification is more common in a mass and intraductal in AB US, and tend to associated with architectural distortion or duct changes. Also, AB US has higher sensitivity than mammography to malignant microcalcification, which is expected to become an effective supplementary examination method for breast microcalcifications, especially in dense breasts.

Two-Phase Speckle Noise Removal in US Images: Speckle Reducing Improved Anisotropic Diffusion and Optimal Bayes Threshold

Medial images are contaminated by multiplicative speckle noise, which dramatically reduces ultrasound images and has a detrimental impact on a variety of image interpretation tasks. Hence, to overcome this issue, this paper presented a Two-Phase Speckle Reduction approach with Improved Anisotropic Diffusion and Optimal Bayes Threshold termed TPSR-IADOT, which includes the phases like image enhancement and two-level decomposition processes. Initially, the speckle noise is subjected to an image enhancement process where the Speckle Reducing Improved Anisotropic Diffusion (SRAID) filtering process is carried out for the speckle removal process. Afterwards, two-level decomposition takes place which utilizes Discrete Wavelet Transform (DWT) to remove the residual noise. As the speckle noise is mostly present in the high-frequency band, Improved Bayes Threshold will be applied to the high- frequency subbands. Finally, to provide the best outcomes, an optimization algorithm termed Self Improved Pelican Optimization Algorithm (SI-POA) in this work via choosing the optimal threshold value. The efficiency of the proposed method has been validated on an ultrasound image database using Simulink in terms of PSNR, SSIM, SDME and MAPE. Accordingly, from the analysis, it is proved that the proposed TPSR-IADOT attains the PSNR of 40.074, whereas the POA is 38.572, COOT is 38.572, BES is 37.003, PRO is 30.419, WOA is 33.218, RFU-LA is 29.935 and SSI-COA is 39.256, for noise variance[Formula: see text]0.1.

Normal morphometry of the biliary tree in pediatric and adult populations: A systematic review and meta-analysis

PurposeThis study aimed to characterise the normal morphometry of the biliary tree in pediatric and adult populations, through a systematic review and meta-analysis. MethodsThis study, conducted using the PRISMA guidelines and registered with PROSPERO, searched MEDLINE, EMBASE, SCOPUS and Web of Science databases up to October 2022, and updated to August 2023. Studies that reported extractable data on diameter and length of the right, left and common hepatic ducts (LHD, RHD and CHD), and common bile duct (CBD) were included. Quality of the included studies were assessed using the Anatomical Quality Assessment (AQUA) tool. Statistical analysis included subgroup analyses according to sex, age, geographical location, and imaging modality. ResultsIn total, 60 studies were included, of which 44 studies reported adequate data for meta-analysis on 23,796 subjects. Overall, the pooled mean diameter of the CBD was 4.69 mm (95 % CI: 4.28–5.11). Significant differences were found between pediatric (1.32 mm, 95 % CI: 1.03–1.61) and adult (4.97 mm, 95 % CI: 4.67–5.27) subjects, as well as US (3.82 mm, 95 % CI: 3.15–4.49) and other imaging modalities, including MRI (6.21 mm, 95 % CI: 4.85–7.57) and ERCP (7.24 mm, 95 % CI: 6.08–8.40). The CBD diameter measured significantly larger distally (5.20 mm, 95 % CI: 4.60–5.80) than proximally (4.01 mm, 95 % CI: 3.51–4.51). ConclusionsThe results obtained from this evidence-based study may guide the establishment of standardised reference values and ranges of the normal biliary tree in pediatric and adult populations and aid clinical understanding.

Explainable tele-ultrasound diagnosis of blast lung injury via incremental multimodal multistage fusion and video transformer

Blast lung injury (BLI) is a significant concern in cases of explosive events and can have severe consequences for victims. The injury is often irreversible, emphasizing the need for early symptom confirmation and intervention. Tele-ultrasound (Tele-US) is a novel approach that enables healthcare professionals to remotely interpret US images and provide diagnostic assessments, which can offer timely and accurate assessments for BLI. In this study, we propose an explainable two-stream tele-US diagnosis method for BLI, which is an incremental multimodal multistage fusion method (IM3S-Fusion). This method is the first deep learning-based diagnosis method for BLI and, specifically, the first exploration of video Transformers in this field. It integrates optical flow into US videos, enhancing model interpretability and improving diagnostic accuracy. By integrating incremental broad learning system (IBLS), IM3S-Fusion can incrementally update its parameters with new samples in a lifelong learning fashion, especially for those misdiagnosed samples. The results show its superiority in diagnosing BLI with (87.50 ± 2.76)%, (89.06 ± 2.71)%, (88.09 ± 5.26)%, (85.71 ± 7.14)% and (88.43 ± 2.24)% in accuracy, recall, precision, specificity and F1-score, respectively. Explanatory assistance including fused US video with optical flow, class activation map videos and confidence scores, is provided for doctors to facilitate their diagnosis.

Comparative Analysis of Examination Methods for Periapical Lesion Diagnostics: Assessing Cone-Beam Computer Tomography, Ultrasound, and Periapical Radiography.

Periapical lesions of teeth are typically evaluated using periapical X-rays (PA) or cone-beam computer tomography (CBCT); however, ultrasound imaging (US) can also be used to detect bone defects. A comparative analysis is necessary to establish the diagnostic accuracy of US for the detection of periapical lesions in comparison with PA and CBCT. This study aimed to evaluate and compare the measurement precision of US against PA and CBCT in detecting periapical lesions. This study included 43 maxillary and mandibular teeth with periapical lesions. All teeth were examined clinically, radiographically, and ultrasonographically. Observers evaluated and measured the periapical lesions on CBCT, PA, and US images. The comparison of lesion size showed that it differs significantly between the different methods of examination. A statistically significant difference was found between CBCT and US (mean difference = 0.99 mm, 95% CI [0.43-1.55]), as well as between CBCT and PA (mean difference = 0.61 mm, 95% CI [0.17-1.05]). No difference was found between the US and PA methods (p = 0.193). US cannot replace PA radiography in detecting pathologies but it can accurately measure and characterize periapical lesions with minimal radiation exposure. CBCT is the most precise and radiation-intensive method so it should only be used for complex cases.

Society of Radiologists in Ultrasound Consensus on Routine Pelvic US for Endometriosis.

Endometriosis is a prevalent and potentially debilitating condition that mostly affects individuals of reproductive age, and often has a substantial diagnostic delay. US is usually the first-line imaging modality used when patients report chronic pelvic pain or have issues of infertility, both common symptoms of endometriosis. Other than the visualization of an endometrioma, sonologists frequently do not appreciate endometriosis on routine transvaginal US images. Given a substantial body of literature describing techniques to depict endometriosis at US, the Society of Radiologists in Ultrasound convened a multidisciplinary panel of experts to make recommendations aimed at improving the screening process for endometriosis. The panel was composed of experts in the imaging and management of endometriosis, including radiologists, sonographers, gynecologists, reproductive endocrinologists, and minimally invasive gynecologic surgeons. A comprehensive literature review combined with a modified Delphi technique achieved a consensus. This statement defines the targeted screening population, describes techniques for augmenting pelvic US, establishes direct and indirect observations for endometriosis at US, creates an observational grading and reporting system, and makes recommendations for additional imaging and patient management. The panel recommends transvaginal US of the posterior compartment, observation of the relative positioning of the uterus and ovaries, and the uterine sliding sign maneuver to improve the detection of endometriosis. These additional techniques can be performed in 5 minutes or less and could ultimately decrease the delay of an endometriosis diagnosis in at-risk patients.

Super-Resolution US Imaging of Focal Nodular Hyperplasia.

Super-Resolution US Imaging of Focal Nodular Hyperplasia.

US-based Sequential Algorithm Integrating an AI Model for Advanced Liver Fibrosis Screening.

Background Noninvasive tests can be used to screen patients with chronic liver disease for advanced liver fibrosis; however, the use of single tests may not be adequate. Purpose To construct sequential clinical algorithms that include a US deep learning (DL) model and compare their ability to predict advanced liver fibrosis with that of other noninvasive tests. Materials and Methods This retrospective study included adult patients with a history of chronic liver disease or unexplained abnormal liver function test results who underwent B-mode US of the liver between January 2014 and September 2022 at three health care facilities. A US-based DL network (FIB-Net) was trained on US images to predict whether the shear-wave elastography (SWE) value was 8.7 kPa or higher, indicative of advanced fibrosis. In the internal and external test sets, a two-step algorithm (Two-step#1) using the Fibrosis-4 Index (FIB-4) followed by FIB-Net and a three-step algorithm (Three-step#1) using FIB-4 followed by FIB-Net and SWE were used to simulate screening scenarios where liver stiffness measurements were not or were available, respectively. Measures of diagnostic accuracy were calculated using liver biopsy as the reference standard and compared between FIB-4, SWE, FIB-Net, and European Association for the Study of the Liver guidelines (ie, FIB-4 followed by SWE), along with sequential algorithms. Results The training, validation, and test data sets included 3067 (median age, 42 years [IQR, 33-53 years]; 2083 male), 1599 (median age, 41 years [IQR, 33-51 years]; 1124 male), and 1228 (median age, 44 years [IQR, 33-55 years]; 741 male) patients, respectively. FIB-Net obtained a noninferior specificity with a margin of 5% (P < .001) compared with SWE (80% vs 82%). The Two-step#1 algorithm showed higher specificity and positive predictive value (PPV) than FIB-4 (specificity, 79% vs 57%; PPV, 44% vs 32%) while reducing unnecessary referrals by 42%. The Three-step#1 algorithm had higher specificity and PPV compared with European Association for the Study of the Liver guidelines (specificity, 94% vs 88%; PPV, 73% vs 64%) while reducing unnecessary referrals by 35%. Conclusion A sequential algorithm combining FIB-4 and a US DL model showed higher diagnostic accuracy and improved referral management for all-cause advanced liver fibrosis compared with FIB-4 or the DL model alone. © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Ghosh in this issue.

A systematic review of machine learning based thyroid tumor characterisation using ultrasonographic images.

Ultrasonography is widely used to screen thyroid tumors because it is safe, easy to use, and low-cost. However, it is simultaneously affected by speckle noise and other artifacts, so early detection of thyroid abnormalities becomes difficult for the radiologist. Therefore, various researchers continuously address the limitations of sonography and improve the diagnosis potential of US images for thyroid tissue from the last three decays. Accordingly, the present study extensively reviewed various CAD systems used to classify thyroid tumor US (TTUS) images related to datasets, despeckling algorithms, segmentation algorithms, feature extraction and selection, assessment parameters, and classification algorithms. After the exhaustive review, the achievements and challenges have been reported, and build a road map for the new researchers.

Assessment of Thyroid Gland in Children with Point-of-Care Ultrasound (POCUS): Radiological Performance and Feasibility of Handheld Ultrasound in Clinical Practice

Point-of-Care Ultrasound (POCUS) refers to the use of portable ultrasound machines to perform quick and focused ultrasound examinations at a patient’s bedside or point-of-care. POCUS can be performed by all health workers with specific training to use POCUS. The aim of this study was to investigate the radiological performance and feasibility of POCUS using a handheld ultrasound device (HHUSD) in children for examining the thyroid gland. A pediatric endocrinologist performed thyroid imaging in children referred to our hospital with suspected thyroid disease using an HHUSD. The same children underwent ultrasonography (US) imaging using the same device by the first radiologist, and a second radiologist performed thyroid US using an advanced high-range ultrasound device (AHUSD) (defined as the gold-standard method) within two hours. The data obtained by the three researchers were compared with each other. This study included 105 patients [68.6% girls (n=72)] with a mean age 12.8±3.6 years. When the thyroid volume was evaluated, a strong correlation was found between the measurements of the three researchers (AA vs. MG: r=0.963, AA vs. GT: r=0.969, MG vs. GT: r=0.963, p<0.001). According to the Bland-Altman analysis for total thyroid volume, AA measured 0.43 cc [95% confidence interval (CI): -0.89-0.03] smaller than MG, and 0.11 cc (95% CI: -0.30-0.52) larger than GT, whereas MG measured 0.52 cc (95% CI: 0.09-0.94) larger than GT. When evaluated for the presence of goiter and nodules, a near-perfect agreement was found between the results of the three researchers (AA vs. GT; κ=0.863, MG vs. GT; κ=0.887, p<0.001, and AA vs. GT; κ=1.000, MG vs. GT; κ=0.972, p<0.001, respectively). When evaluated in terms of the longest axis of nodules, a high correlation was found between the measurements of the three researchers (AA vs. MG; r=0.993, AA vs. GT; r=0.996, MG vs. GT; r=0.996, p<0.001). When evaluated in terms of the final diagnosis, the evaluations of the three researchers showed excellent agreement with each other (AA vs. GT; κ=0.893, MG vs. GT; κ=0.863, p<0.001, accuracy rate AA vs. GT: 93.3%; MG vs. GT: 91.4%). A pediatric endocrinologist, equipped with sufficient training in thyroid US evaluation, incorporated HHUSD examination as a routine clinical tool in an outpatient setting. It was shown that, they could effectively assess normal thyroid tissue in pediatric patients. Moreover, the HHUSD proved to be useful in detecting thyroid pathologies. However, it is important to note that for a more comprehensive evaluation of thyroid nodules, including detailed assessment and Thyroid Imaging Reporting and Data System (TIRADS) classification, patients should be referred to radiology departments equipped with AHUSD systems. These specialized devices, along with the expertise of radiologists, are essential for in-depth evaluations and accurate classification of thyroid nodules.

Abstract 68: Imaging Endothelial PD-L1 expression with ultrasound molecular imaging to stratify immune checkpoint inhibitor responders

Abstract Immune checkpoint blockade therapies (ICTs) have revolutionized oncology through substantial and lasting treatment outcomes, but only in 15-20% of patients. Non-invasive diagnostic tools to select patients that are most likely to respond to ICT and guide treatment regimens, and thus minimize toxicities to non-responder patients, are in high demand. Program death-1 ligand (PD-L1) marker evaluation in the tumor vasculature and tumor microenvironment is an essential criterion for identifying a suitable patient for PD-L1 based ICT, and it also serves as a prognostic marker in tumor response evaluation. Ultrasound molecular imaging (USMI) is an up-and-coming non-invasive imaging modality that enables longitudinal characterization of endothelial-specific markers. We are developing USMI methods to image immune markers expressed specifically on endothelial cells (EC) of tumors as a decision-support tool for guiding ICT strategies. In this work, we demonstrate for the first time the potential of imaging EC-specific immunosuppressive markers such as PD-L1 and investigate whether it can act as a marker for predicting ICT response. Target-ready microbubble (MB) was used (Trust Bio-sonics, Taiwan) to generate PD-L1 antibody (Ab) linked MBs (TMBs). DSPE-PEG was functionalized to anti-PD-L1 Ab or isotype-Ab (control) to generate stable TMBs (~5 × 107 MBs/mL) to target PD-L1 expressing ECs in vivo. We studied the treatment effect of anti-PD-L1-Ab and isotype-Ab using a syngeneic CT-26 tumor in mice and demonstrated the targeting potency of TMB or iso-Ab-MB (NMB). US imaging was performed using Vevo 2100 high-frequency US system with M250 transducer to evaluate and quantify the differentially targeted enhancement parameter (dTE). We observed that TMBs bound PD-L1 expressing ECs in vivo, with high and low enhancement for TMB and NMB, respectively, in tumors validated by dTE quantification through characterization of the slower washout kinetics. Furthermore, TMBs exhibited increased tumor retention and prolonged binding with average differential targeted enhancement values (dTE) (+/- SD) of 5.2 ± 7.4 for TMBs compared to 2.4 ± 3.9 for NMBs (p=0.14). Preliminary results of USMI (dTE values of TMBs) indicate that PD-L1-treated mice showed a higher response (4.5 ± 5.2; n=9) to the treatment group than control mice that did not receive Ab (1.4 ± 0.8; n=4). The tumor growth rate for the treated mice group was 5.4 -fold retarded compared to the control group which correlated with USMI dTE values. The USMI signals correlated with IHC performed on tumor tissue ex vivo from the respective mouse used for US imaging. Overall, our pre-clinical study suggests that USMI using TMBs may contribute to non-invasive assessment of IC expression on tumor ECs, and that it may serve as a predictive biomarker of ICT response. This novel TMB agent could be a game changer for CE USMI-based cancer prognosis in ICB therapy. Citation Format: Arutselvan Natarajan, Farbod Tabesh, Mahsa Bataghva, Ramasamy Paulmurugan, Ahmed Nagy El Kaffas. Imaging Endothelial PD-L1 expression with ultrasound molecular imaging to stratify immune checkpoint inhibitor responders [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 68.

NTRK-rearranged spindle cell neoplasm: Initial observation of imaging appearance and clinicopathologic correlation

ObjectiveTo explore pre-treatment imaging findings of neurotrophic tyrosine receptor kinase (NTRK)-rearranged spindle cell neoplasm, an emerging group of molecularly defined soft tissue tumors and summarize the clinical course, including TRK inhibitor therapy response. Materials and methodsThis retrospective study included 8 women and 4 men with NTRK-rearranged spindle cell neoplasm (median age, 35.5 years, range, 0–66). Available pre-treatment MRI, CT, PET, and US imaging were reviewed. Tumor histology and the patients' clinical course were reviewed. ResultsPrimary tumors were located within the soft tissue, lungs, kidney, and breast with soft tissue being the most prevalent site (n = 6). Pre-treatment MRI (n = 4) revealed linear hypointense signal foci and contrast enhancement in all patients with hemorrhage in half of the tumors. A tail sign (n = 1) and fluid levels (n = 1) were less frequent. Ultrasound showed well-marginated hypoechoic masses with internal flow. Primary tumors were all non-calcified on CT (4/4). Metastases were FDG-avid (4/4). Among the 8 patients who developed metastasis, 7 developed pulmonary metastases. All four patients who received NTRK inhibitor therapy showed an initial decrease in tumor size or FDG uptake. ConclusionNTRK-rearranged neoplasms may occur as enhancing masses with linear hypointense signal foci on MRI and FDG avid metastases on PET. Pulmonary metastases were frequent in our study. Initial treatment response is observed in most patients.

Automatic hepatic tumor segmentation in intra-operative ultrasound: a supervised deep-learning approach.

Training and evaluation of the performance of a supervised deep-learning model for the segmentation of hepatic tumors from intraoperative US (iUS) images, with the purpose of improving the accuracy of tumor margin assessment during liver surgeries and the detection of lesions during colorectal surgeries. In this retrospective study, a U-Net network was trained with the nnU-Net framework in different configurations for the segmentation of CRLM from iUS. The model was trained on B-mode intraoperative hepatic US images, hand-labeled by an expert clinician. The model was tested on an independent set of similar images. The average age of the study population was 61.9 ± 9.9 years. Ground truth for the test set was provided by a radiologist, and three extra delineation sets were used for the computation of inter-observer variability. The presented model achieved a DSC of 0.84 (), which is comparable to the expert human raters scores. The model segmented hypoechoic and mixed lesions more accurately (DSC of 0.89 and 0.88, respectively) than hyper- and isoechoic ones (DSC of 0.70 and 0.60, respectively) only missing isoechoic or >20mm in diameter (8% of the tumors) lesions. The inclusion of extra margins of probable tumor tissue around the lesions in the training ground truth resulted in lower DSCs of 0.75 (). The model can accurately segment hepatic tumors from iUS images and has the potential to speed up the resection margin definition during surgeries and the detection of lesion in screenings by automating iUS assessment.

Deep learning for synovial volume segmentation of the first carpometacarpal joint in osteoarthritis patients

ObjectiveThe objective of this study was to develop a deep-learning-based approach to automatically segment 3D ultrasound images of the synovial tissue in osteoarthritis of the first carpometacarpal (CMC1 OA). DesignDeep learning predictions on 2D ultrasound slices sampled in the transverse plane were used to view the synovial tissue of the first carpometacarpal in patients with OA, followed by reconstruction into 3D surfaces. A modified 2D U-Net was trained using a dataset of 832 2D US images resliced from 89 3D US images. Segmentation accuracy was evaluated using a testing dataset of 208 2D US images resliced from 15 3D US images. Absolute and signed performance metrics were computed, and segmentation performance was compared between the manual segmentations of raters 1 and 2. ResultsResults of the U-Net-based run were mean 3D DSC 86.9 ± 4.8%, recall 93.7 ± 3.6%, precision 81.1 ± 6.9%, volume percent difference 16.9 ± 10.2%, mean surface distance 0.18 ± 0.04 mm, and Hausdorff distance 1.8 ± 0.8 mm. The algorithm demonstrated an overall increase in performance after 3D segmentation reconstruction compared to 2D predictions, but the difference was not statistically significant. ConclusionThis study investigated the use of a modified U-Net algorithm to automatically segment the synovial tissue volume (STV) of CMC1 OA patients and demonstrated that the addition of this deep learning method increases the efficiency of STV estimations in clinical trial settings.

Collaborative Enhancement of Consistency and Accuracy in US Diagnosis of Thyroid Nodules Using Large Language Models.

Background Large language models (LLMs) hold substantial promise for medical imaging interpretation. However, there is a lack of studies on their feasibility in handling reasoning questions associated with medical diagnosis. Purpose To investigate the viability of leveraging three publicly available LLMs to enhance consistency and diagnostic accuracy in medical imaging based on standardized reporting, with pathology as the reference standard. Materials and Methods US images of thyroid nodules with pathologic results were retrospectively collected from a tertiary referral hospital between July 2022 and December 2022 and used to evaluate malignancy diagnoses generated by three LLMs-OpenAI's ChatGPT 3.5, ChatGPT 4.0, and Google's Bard. Inter- and intra-LLM agreement of diagnosis were evaluated. Then, diagnostic performance, including accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC), was evaluated and compared for the LLMs and three interactive approaches: human reader combined with LLMs, image-to-text model combined with LLMs, and an end-to-end convolutional neural network model. Results A total of 1161 US images of thyroid nodules (498 benign, 663 malignant) from 725 patients (mean age, 42.2 years ± 14.1 [SD]; 516 women) were evaluated. ChatGPT 4.0 and Bard displayed substantial to almost perfect intra-LLM agreement (κ range, 0.65-0.86 [95% CI: 0.64, 0.86]), while ChatGPT 3.5 showed fair to substantial agreement (κ range, 0.36-0.68 [95% CI: 0.36, 0.68]). ChatGPT 4.0 had an accuracy of 78%-86% (95% CI: 76%, 88%) and sensitivity of 86%-95% (95% CI: 83%, 96%), compared with 74%-86% (95% CI: 71%, 88%) and 74%-91% (95% CI: 71%, 93%), respectively, for Bard. Moreover, with ChatGPT 4.0, the image-to-text-LLM strategy exhibited an AUC (0.83 [95% CI: 0.80, 0.85]) and accuracy (84% [95% CI: 82%, 86%]) comparable to those of the human-LLM interaction strategy with two senior readers and one junior reader and exceeding those of the human-LLM interaction strategy with one junior reader. Conclusion LLMs, particularly integrated with image-to-text approaches, show potential in enhancing diagnostic medical imaging. ChatGPT 4.0 was optimal for consistency and diagnostic accuracy when compared with Bard and ChatGPT 3.5. © RSNA, 2024 Supplemental material is available for this article.

Curated and Annotated Dataset of Lung US Images in Zambian Children with Clinical Pneumonia.

See also the commentary by Sitek in this issue. Supplemental material is available for this article.