Automation of Lung Ultrasound Interpretation via Deep Learning for the Classification of Normal versus Abnormal Lung Parenchyma: A Multicenter Study.
Ontology highlight
ABSTRACT:
SUBMITTER: Arntfield R
PROVIDER: S-EPMC8621216 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:Prediction of kidney function and chronic kidney disease (CKD) through kidney ultrasound imaging has long been considered desirable in clinical practice because of its safety, convenience, and affordability. However, this highly desirable approach is beyond the capability of human vision. We developed a deep learning approach for automatically determining the estimated glomerular filtration rate (eGFR) and CKD status. We exploited the transfer learning technique, integrating the powerful ResNet model pretrained on an ImageNet dataset in our neural network architecture, to predict kidney function based on 4,505 kidney ultrasound images labeled using eGFRs derived from serum creatinine concentrations. To further extract the information from ultrasound images, we leveraged kidney length annotations to remove the peripheral region of the kidneys and applied various data augmentation schemes to produce additional data with variations. Bootstrap aggregation was also applied to avoid overfitting and improve the model's generalization. Moreover, the kidney function features obtained by our deep neural network were used to identify the CKD status defined by an eGFR of <60?ml/min/1.73?m2. A Pearson correlation coefficient of 0.741 indicated the strong relationship between artificial intelligence (AI)- and creatinine-based GFR estimations. Overall CKD status classification accuracy of our model was 85.6% -higher than that of experienced nephrologists (60.3%-80.1%). Our model is the first fundamental step toward realizing the potential of transforming kidney ultrasound imaging into an effective, real-time, distant screening tool. AI-GFR estimation offers the possibility of noninvasive assessment of kidney function, a key goal of AI-powered functional automation in clinical practice.
| S-EPMC6550224 | biostudies-literature
Project description:The Covid-19 European outbreak in February 2020 has challenged the world's health systems, eliciting an urgent need for effective and highly reliable diagnostic instruments to help medical personnel. Deep learning (DL) has been demonstrated to be useful for diagnosis using both computed tomography (CT) scans and chest X-rays (CXR), whereby the former typically yields more accurate results. However, the pivoting function of a CT scan during the pandemic presents several drawbacks, including high cost and cross-contamination problems. Radiation-free lung ultrasound (LUS) imaging, which requires high expertise and is thus being underutilised, has demonstrated a strong correlation with CT scan results and a high reliability in pneumonia detection even in the early stages. In this study, we developed a system based on modern DL methodologies in close collaboration with Fondazione IRCCS Policlinico San Matteo's Emergency Department (ED) of Pavia. Using a reliable dataset comprising ultrasound clips originating from linear and convex probes in 2908 frames from 450 hospitalised patients, we conducted an investigation into detecting Covid-19 patterns and ranking them considering two severity scales. This study differs from other research projects by its novel approach involving four and seven classes. Patients admitted to the ED underwent 12 LUS examinations in different chest parts, each evaluated according to standardised severity scales. We adopted residual convolutional neural networks (CNNs), transfer learning, and data augmentation techniques. Hence, employing methodological hyperparameter tuning, we produced state-of-the-art results meeting F1 score levels, averaged over the number of classes considered, exceeding 98%, and thereby manifesting stable measurements over precision and recall.
| S-EPMC8349313 | biostudies-literature
Project description:2 normal lung parenchyma SAGE libraries, generated from 2 pools of 4 individuals each Discussed in the study: IDENTIFICATION OF NOVEL LUNG GENES IN BRONCHIAL EPITHELIUM BY SERIAL ANALYSIS OF GENE EXPRESSION Kim M. Lonergan1, Raj Chari1, Ronald J. deLeeuw1, Ashleen Shadeo1, Bryan Chi1, Ming-Sound Tsao2, Steven Jones3, Marco Marra3, Victor Ling1, Raymond Ng1,4, Calum MacAulay5, Stephen Lam5 and Wan L. Lam1 From the 1Department of Cancer Genetics & Developmental Biology, 5Department of Cancer Imaging, 3Canada's Michael Smith Genome Sciences Centre, British Columbia Cancer Research Centre, Vancouver, BC, Canada, 2Ontario Cancer Institute / Princess Margaret Hospital, Toronto, ON, Canada, 4the Department of Computer Science, University of British Columbia, Vancouver, BC, Canada A description of the transcriptome of human bronchial epithelium should provide a basis for studying lung diseases including cancer. We demonstrate here that minute epithelial specimens obtained by bronchial brushings afford reliable profiling by serial analysis of gene expression (SAGE) leading to lung gene discovery. We have deduced global gene expression profiles of bronchial epithelium and lung parenchyma, based upon a vast data set of nearly two million sequence tags from 21 SAGE libraries generated from individuals with a history of smoking. Cluster and linear regression analysis demonstrate the repeatability and reproducibility of bronchial SAGE libraries, and suggest that the transcriptome of the bronchial epithelium is distinct from that of lung parenchyma and other tissue types. This distinction is highlighted by the abundant expression of signature genes that reflect tissue-specific and region-specific functions. Through our analysis we have identified novel bronchial-enriched genes and a novel transcript variant for surfactant, pulmonary-associated protein B in lung parenchyma. Conspicuously, gene expression associated with ciliogenesis is evident in bronchial epithelium. Additionally, it is noted that a large number of unmapped tags awaits further investigation. This study represents a comprehensive delineation of the bronchial and parenchyma transcriptomes, identifying more than 20,000 known and hypothetical genes expressed in the human lung, constituting one of the largest human SAGE studies reported to date. Keywords: lung parenchyma 2 normal lung parenchyma SAGE libraries, generated from 2 pools of 4 individuals each
2006-11-20 | E-GEOD-3708 | biostudies-arrayexpress
Project description:2 normal lung parenchyma SAGE libraries, generated from 2 pools of 4 individuals each Discussed in the study: IDENTIFICATION OF NOVEL LUNG GENES IN BRONCHIAL EPITHELIUM BY SERIAL ANALYSIS OF GENE EXPRESSION Kim M. Lonergan1, Raj Chari1, Ronald J. deLeeuw1, Ashleen Shadeo1, Bryan Chi1, Ming-Sound Tsao2, Steven Jones3, Marco Marra3, Victor Ling1, Raymond Ng1,4, Calum MacAulay5, Stephen Lam5 and Wan L. Lam1 From the 1Department of Cancer Genetics & Developmental Biology, 5Department of Cancer Imaging, 3Canada's Michael Smith Genome Sciences Centre, British Columbia Cancer Research Centre, Vancouver, BC, Canada, 2Ontario Cancer Institute / Princess Margaret Hospital, Toronto, ON, Canada, 4the Department of Computer Science, University of British Columbia, Vancouver, BC, Canada A description of the transcriptome of human bronchial epithelium should provide a basis for studying lung diseases including cancer. We demonstrate here that minute epithelial specimens obtained by bronchial brushings afford reliable profiling by serial analysis of gene expression (SAGE) leading to lung gene discovery. We have deduced global gene expression profiles of bronchial epithelium and lung parenchyma, based upon a vast data set of nearly two million sequence tags from 21 SAGE libraries generated from individuals with a history of smoking. Cluster and linear regression analysis demonstrate the repeatability and reproducibility of bronchial SAGE libraries, and suggest that the transcriptome of the bronchial epithelium is distinct from that of lung parenchyma and other tissue types. This distinction is highlighted by the abundant expression of signature genes that reflect tissue-specific and region-specific functions. Through our analysis we have identified novel bronchial-enriched genes and a novel transcript variant for surfactant, pulmonary-associated protein B in lung parenchyma. Conspicuously, gene expression associated with ciliogenesis is evident in bronchial epithelium. Additionally, it is noted that a large number of unmapped tags awaits further investigation. This study represents a comprehensive delineation of the bronchial and parenchyma transcriptomes, identifying more than 20,000 known and hypothetical genes expressed in the human lung, constituting one of the largest human SAGE studies reported to date. Keywords: lung parenchyma
2006-11-21 | GSE3708 | GEO
Project description:Read-through transcripts result from the continuous transcription of adjacent, similarly oriented genes, with the splicing out of the intergenic region. They have been found in several neoplastic and normal tissues, but their pathophysiological significance is unclear. We used high-throughput sequencing of cDNA fragments (RNA-Seq) to identify read-through transcripts in the non-involved lung tissue of 64 surgically treated lung adenocarcinoma patients. A total of 52 distinct read-through species was identified, with 24 patients having at least one read-through event, up to a maximum of 17 such transcripts in one patient. Sanger sequencing validated 28 of these transcripts and identified an additional 15, for a total of 43 distinct read-through events involving 35 gene pairs. Expression levels of 10 validated read-through transcripts were measured by quantitative PCR in pairs of matched non-involved lung tissue and lung adenocarcinoma tissue from 45 patients. Higher expression levels were observed in normal lung tissue than in the tumor counterpart, with median relative quantification ratios between normal and tumor varying from 1.90 to 7.78; the difference was statistically significant (P < 0.001, Wilcoxon's signed-rank test for paired samples) for eight transcripts: ELAVL1-TIMM44, FAM162B-ZUFSP, IFNAR2-IL10RB, INMT-FAM188B, KIAA1841-C2orf74, NFATC3-PLA2G15, SIRPB1-SIRPD, and SHANK3-ACR. This report documents the presence of read-through transcripts in apparently normal lung tissue, with inter-individual differences in patterns and abundance. It also shows their down-regulation in tumors, suggesting that these chimeric transcripts may function as tumor suppressors in lung tissue.
| S-EPMC5053695 | biostudies-literature
Project description:ObjectivesTo evaluate the feasibility and reproducibility of superb microvascular imaging (SMI) of the neonatal brain and to describe normal imaging features.MethodsWe performed transcranial ultrasound with SMI in 19 healthy term-born neonates. SMI was done according to a structured examination protocol, using two linear 18 MHz and 14 MHz transducers. Superficial and deep scans were acquired in the coronal and sagittal planes, using the left and right superior frontal gyri as anatomical landmarks. All SMI views were imaged by monochrome and colour SMI and evaluated with respect to visibility of extrastriatal (i.e. cortical and medullary) and striatal microvessels.ResultsWe have described normal morphologic features of intraparenchymal brain microvasculature as "short parallel" cortical vessels, "smoothly curved" medullary vessels, and deep striatal vessels. In general, SMI performance was better on coronal views than on sagittal views. On superficial coronal scans, cortical microvessels were identifiable in 90-100%, medullary microvessels in 95-100%. On deep scans, cortical and medullary microvessels were visible in all cases, while striatal microvessels were identifiable in 71% of cases.ConclusionsCerebral SMI ultrasound is feasible and well-reproducible and provides a novel non-invasive imaging tool for the assessment of intraparenchymal brain microvasculature (extrastriatal and striatal microvessels) in neonates without the use of contrast.Key points• Superb microvascular imaging (SMI) of the neonatal brain is feasible and reproducible. • SMI depicts extrastriatal and striatal microvessels. • SMI detects two types of extrastriatal microvessels: cortical and medullary.
| S-EPMC6420458 | biostudies-literature
Project description:OBJECTIVE: To demonstrate normal and abnormal findings of distal brachialis tendon attachment in cadavers, normal volunteers and patients by means of ultrasound. METHODS: 3 cadaveric specimens, 30 normal volunteers and 125 patients were evaluated by means of ultrasound. Correlative MRI was obtained in volunteers. RESULTS: In all cases, ultrasound demonstrated the distal brachialis tendon shaped by two distinct tendons belonging to the deep head and superficial head of the brachialis muscle. Correlative MRI demonstrated that the brachialis is composed of two distinct tendons in 83% of volunteers (25/30). In the patient group, four avulsions with bony detachment involving the deep head, one delayed onset muscular soreness and three tendinous detachments with no bony avulsion involving one or two tendons were identified. The four patients with bony avulsion were immediately referred to the orthopaedic surgeon for a pre-surgical evaluation. Patients without bony avulsion were not referred to the surgeon. CONCLUSION: Detailed anatomy of the distal tendon is discernible in 100% of cases with ultrasound. There are two distinct tendons, and ultrasound can differentiate isolated lesions. In patients with distal brachialis tendon lesions, ultrasound may modify the clinical management of the patient. Advances in knowledge Detailed anatomy of the distal brachialis tendon is discernible with ultrasound and there are two distinct tendons.
| S-EPMC3635801 | biostudies-other
Project description:This study aimed to assess the diagnostic performance of computed tomography (CT) for differentiating diffuse thyroid disease (DTD) from normal thyroid parenchyma (NTP) using multicenter data. Between January 2016 and June 2016, 229 patients underwent preoperative neck CT and subsequent thyroid surgery at five participating institutions. The neck CT images of each patient were retrospectively reviewed and classified into the following four categories: no DTD, indeterminate, suspicious for DTD, and DTD. The results of the CT image evaluations were compared with the histopathological results to determine the diagnostic accuracy of CT at each institution. According to the histopathological results, there were NTP (n = 151), Hashimoto thyroiditis (n = 24), non-Hashimoto lymphocytic thyroiditis (n = 47), and diffuse hyperplasia (n = 7). The CT categories of the 229 patients were "no DTD" in 89 patients, "indeterminate" in 40 patients, "suspicious for DTD" in 42 patients, and "DTD" in 58 patients. The presence of two or more CT features of DTD, which was classified as "suspicious for DTD" by all radiologists, had the largest area under the receiver-operating characteristic curve (Az = 0.820; 95% confidence interval: 0.764, 0.868), with sensitivity of 85.9% and specificity of 78.2%. However, no statistical significance between readers' experience and their diagnostic accuracy was found. In conclusion, evaluations of CT images are helpful for differentiating DTD from NTP.
| S-EPMC6237331 | biostudies-literature
Project description:The treatment of brain diseases is hindered by the blood-brain barrier (BBB) preventing most drugs from entering the brain. Focused ultrasound (FUS) with microbubbles can open the BBB safely and reversibly. Systemic drug injection might induce toxicity, but encapsulation into nanoparticles reduces accumulation in normal tissue. Here we used a novel platform based on poly(2-ethyl-butyl cyanoacrylate) nanoparticle-stabilized microbubbles to permeabilize the BBB in a melanoma brain metastasis model. With a dual-frequency ultrasound transducer generating FUS at 1.1 MHz and 7.8 MHz, we opened the BBB using nanoparticle-microbubbles and low-frequency FUS, and applied high-frequency FUS to generate acoustic radiation force and push nanoparticles through the extracellular matrix. Using confocal microscopy and image analysis, we quantified nanoparticle extravasation and distribution in the brain parenchyma. We also evaluated haemorrhage, as well as the expression of P-glycoprotein, a key BBB component. FUS and microbubbles distributed nanoparticles in the brain parenchyma, and the distribution depended on the extent of BBB opening. The results from acoustic radiation force were not conclusive, but in a few animals some effect could be detected. P-glycoprotein was not significantly altered immediately after sonication. In summary, FUS with our nanoparticle-stabilized microbubbles can achieve accumulation and displacement of nanoparticles in the brain parenchyma.
| S-EPMC5770053 | biostudies-literature
Project description:With many thyroid nodules being incidentally detected, it is important to identify as many malignant nodules as possible while excluding those that are highly likely to be benign from fine needle aspiration (FNA) biopsies or surgeries. This paper presents a computer-aided diagnosis (CAD) system for classifying thyroid nodules in ultrasound images. We use deep learning approach to extract features from thyroid ultrasound images. Ultrasound images are pre-processed to calibrate their scale and remove the artifacts. A pre-trained GoogLeNet model is then fine-tuned using the pre-processed image samples which leads to superior feature extraction. The extracted features of the thyroid ultrasound images are sent to a Cost-sensitive Random Forest classifier to classify the images into "malignant" and "benign" cases. The experimental results show the proposed fine-tuned GoogLeNet model achieves excellent classification performance, attaining 98.29% classification accuracy, 99.10% sensitivity and 93.90% specificity for the images in an open access database (Pedraza et al. 16), while 96.34% classification accuracy, 86% sensitivity and 99% specificity for the images in our local health region database.
| S-EPMC5537102 | biostudies-other