Automatic Deep Learning-Based Pipeline for Automatic Delineation and Measurement of Fetal Brain Structures in Routine Mid-Trimester Ultrasound Images

INTRODUCTION

The aim of this study was to develop a pipeline using state-of-the-art deep learning methods to automatically delineate and measure several of the most important brain structures in fetal brain ultrasound (US) images.

METHODS

The dataset was composed of 5,331 images of the fetal brain acquired during the routine mid-trimester US scan. Our proposed pipeline automatically performs the following three steps: brain plane classification (transventricular, transthalamic, or transcerebellar plane); brain structures delineation (9 different structures); and automatic measurement (from the structure delineations). The methods were trained on a subset of 4,331 images and each step was evaluated on the remaining 1,000 images.

RESULTS

Plane classification reached 98.6% average class accuracy. Brain structure delineation obtained an average pixel accuracy higher than 96% and a Jaccard index higher than 70%. Automatic measurements get an absolute error below 3.5% for the four standard head biometries (head circumference, biparietal diameter, occipitofrontal diameter, and cephalic index), 9% for transcerebellar diameter, 12% for cavum septi pellucidi ratio, and 26% for Sylvian fissure operculization degree.

CONCLUSIONS

The proposed pipeline shows the potential of deep learning methods to delineate fetal head and brain structures and obtain automatic measures of each anatomical standard plane acquired during routine fetal US examination.

Noninvasive prediction models of intra-amniotic infection in women with preterm labor

BACKGROUND

Among women with preterm labor, those with intra-amniotic infection present the highest risk of early delivery and the most adverse outcomes. The identification of intra-amniotic infection requires amniocentesis, perceived as too invasive by women and physicians. Noninvasive methods for identifying intra-amniotic infection and/or early delivery are crucial to focus early efforts on high-risk preterm labor women while avoiding unnecessary interventions in low-risk preterm labor women.

OBJECTIVE

This study modeled the best performing models, integrating biochemical data with clinical and ultrasound information to predict a composite outcome of intra-amniotic infection and/or spontaneous delivery within 7 days.

STUDY DESIGN

From 2015 to 2020, data from a cohort of women, who underwent amniocentesis to rule in or rule out intra-amniotic infection or inflammation, admitted with a diagnosis of preterm labor at <34 weeks of gestation at the Hospital Clinic and Hospital Sant Joan de Déu, Barcelona, Spain, were used. At admission, transvaginal ultrasound was performed, and maternal blood and vaginal samples were collected. Using high-dimensional biology, vaginal proteins (using multiplex immunoassay), amino acids (using high-performance liquid chromatography), and bacteria (using 16S ribosomal RNA gene amplicon sequencing) were explored to predict the composite outcome. We selected ultrasound, maternal blood, and vaginal predictors that could be tested with rapid diagnostic techniques and developed prediction models employing machine learning that was applied in a validation cohort.

RESULTS

A cohort of 288 women with preterm labor at <34 weeks of gestation, of which 103 (35%) had a composite outcome of intra-amniotic infection and/or spontaneous delivery within 7 days, were included in this study. The sample was divided into derivation (n=116) and validation (n=172) cohorts. Of note, 4 prediction models were proposed, including ultrasound transvaginal cervical length, maternal C-reactive protein, vaginal interleukin 6 (using an automated immunoanalyzer), vaginal pH (using a pH meter), vaginal lactic acid (using a reflectometer), and vaginal Lactobacillus genus (using quantitative polymerase chain reaction), with areas under the receiving operating characteristic curve ranging from 82.2% (95% confidence interval, ±3.1%) to 85.2% (95% confidence interval, ±3.1%), sensitivities ranging from 76.1% to 85.9%, and specificities ranging from 75.2% to 85.1%.

CONCLUSION

The study results have provided proof of principle of how noninvasive methods suitable for point-of-care systems can select high-risk cases among women with preterm labor and might substantially aid in clinical management and outcomes while improving the use of resources and patient experience.

Prediction of Neonatal Respiratory Morbidity Assessed by Quantitative Ultrasound Lung Texture Analysis in Twin Pregnancies

The objective of this study was to evaluate the performance of quantitative ultrasound of fetal lung texture analysis in predicting neonatal respiratory morbidity (NRM) in twin pregnancies. This was an ambispective study involving consecutive cases. Eligible cases included twin pregnancies between 27.0 and 38.6 weeks of gestation, for which an ultrasound image of the fetal thorax was obtained within 48 h of delivery. Images were analyzed using quantusFLM^® version 3.0. The primary outcome of this study was neonatal respiratory morbidity, defined as the occurrence of either transient tachypnea of the newborn or respiratory distress syndrome. The performance of quantusFLM^® in predicting NRM was analyzed by matching quantitative ultrasound analysis and clinical outcomes. This study included 166 images. Neonatal respiratory morbidity occurred in 12.7% of cases, and it was predicted by quantusFLM^® analysis with an overall sensitivity of 42.9%, specificity of 95.9%, positive predictive value of 60%, and negative predictive value of 92.1%. The accuracy was 89.2%, with a positive likelihood ratio of 10.4, and a negative likelihood ratio of 0.6. The results of this study demonstrate the good prediction capability of NRM in twin pregnancies using a non-invasive lung texture analysis software. The test showed an overall good performance with high specificity, negative predictive value, and accuracy.

Quantitative ultrasound image analysis of axillary lymph nodes to differentiate malignancy from reactive benign changes due to COVID-19 vaccination

Purpose

The aim of this study is to assess the potential of quantitative image analysis and machine learning techniques to differentiate between malignant lymph nodes and benign lymph nodes affected by reactive changes due to COVID-19 vaccination.

Method

In this institutional review board–approved retrospective study, we improved our previously published artificial intelligence model, by retraining it with newly collected images and testing its performance on images containing benign lymph nodes affected by COVID-19 vaccination. All the images were acquired and selected by specialized breast-imaging radiologists and the nature of each node (benign or malignant) was assessed through a strict clinical protocol using ultrasound-guided biopsies.

Results

A total of 180 new images from 154 different patients were recruited: 71 images (10 cases and 61 controls) were used to retrain the old model and 109 images (36 cases and 73 controls) were used to evaluate its performance. The achieved accuracy of the proposed method was 92.7% with 77.8% sensitivity and 100% specificity at the optimal cut-off point. In comparison, the visual node inspection made by radiologists from ultrasound images reached 69.7% accuracy with 41.7% sensitivity and 83.6% specificity.

Conclusions

The results obtained in this study show the potential of the proposed techniques to differentiate between malignant lymph nodes and benign nodes affected by reactive changes due to COVID-19 vaccination. These techniques could be useful to non-invasively diagnose lymph node status in patients with suspicious reactive nodes, although larger multicenter studies are needed to confirm and validate the results.

Concordance of the risk of neonatal respiratory morbidity assessed by quantitative ultrasound lung texture analysis in fetuses of twin pregnancies

To evaluate the concordance of the risk of neonatal respiratory morbidity (NRM) assessed by quantitative ultrasound lung texture analysis (QuantusFLM) between twin fetuses of the same pregnancy. Prospective study conducted in twin pregnancies. Fetal ultrasound lung images were obtained at 26.0–38.6 weeks of gestation. Categorical (high or low) and continuous results of the risk of NRM were compared between twins. Fetal ultrasound lung images from 131 pairs (262 images) of twins were included. The images were classified into three gestational age ranges: Group 1 (26.0–29.6 weeks, 78 images, 39 pairs [29.8%]); Group 2 (30.0–33.6 weeks, 98 images, 49 pairs [37.4%]) and Group 3 (34.0–38.6 weeks, 86 images, 43 pairs [32.8%]). Concordance was good in Groups 1 and 3 and moderate in Group 2. In Groups 2 and 3 at least one fetus presented high-risk results in 26.5% and 11.6% of twin pairs, respectively. Only gestational age < 32 weeks, gestational diabetes mellitus, and spontaneous conception were associated with a high risk of NRM in Group 2. There was good concordance of the risk of NRM between twins < 30.0 weeks and > 34.0 weeks. From 30.0 to 33.6 weeks 26.5% of the twin pairs had discordant results, with moderate concordance of the risk of NRM.

Analysis of maturation features in fetal brain ultrasound via artificial intelligence for the estimation of gestational age

BACKGROUND

Optimal prenatal care relies on accurate gestational age dating. After the first trimester, the accuracy of current gestational age estimation methods diminishes with increasing gestational age. Considering that, in many countries, access to first trimester crown rump length is still difficult owing to late booking, infrequent access to prenatal care, and unavailability of early ultrasound examination, the development of accurate methods for gestational age estimation in the second and third trimester of pregnancy remains an unsolved challenge in fetal medicine.

OBJECTIVE

This study aimed to evaluate the performance of an artificial intelligence method based on automated analysis of fetal brain morphology on standard cranial ultrasound sections to estimate the gestational age in second and third trimester fetuses compared with the current formulas using standard fetal biometry.

STUDY DESIGN

Standard transthalamic axial plane images from a total of 1394 patients undergoing routine fetal ultrasound were used to develop an artificial intelligence method to automatically estimate gestational age from the analysis of fetal brain information. We compared its performance-as stand alone or in combination with fetal biometric parameters-against 4 currently used fetal biometry formulas on a series of 3065 scans from 1992 patients undergoing second (n=1761) or third trimester (n=1298) routine ultrasound, with known gestational age estimated from crown rump length in the first trimester.

RESULTS

Overall, 95% confidence interval of the error in gestational age estimation was 14.2 days for the artificial intelligence method alone and 11.0 when used in combination with fetal biometric parameters, compared with 12.9 days of the best method using standard biometrics alone. In the third trimester, the lower 95% confidence interval errors were 14.3 days for artificial intelligence in combination with biometric parameters and 17 days for fetal biometrics, whereas in the second trimester, the 95% confidence interval error was 6.7 and 7, respectively. The performance differences were even larger in the small-for-gestational-age fetuses group (14.8 and 18.5, respectively).

CONCLUSION

An automated artificial intelligence method using standard sonographic fetal planes yielded similar or lower error in gestational age estimation compared with fetal biometric parameters, especially in the third trimester. These results support further research to improve the performance of these methods in larger studies.

Mid-trimester prediction of spontaneous preterm birth with automated cervical quantitative ultrasound texture analysis and cervical length: a prospective study

The objective of this study was to evaluate a novel automated test based on ultrasound cervical texture analysis to predict spontaneous Preterm Birth (sPTB) alone and in combination with Cervical Length (CL). General population singleton pregnancies between 18 + 0 and 24 + 6 weeks’ gestation were assessed prospectively at two centers. Cervical ultrasound images were evaluated and the occurrence of sPTB before weeks 37 + 0 and 34 + 0 were recorded. CL was measured on-site. The automated texture analysis test was applied offline to all images. Their performance to predict the occurrence of sPTB before 37 + 0 and 34 + 0 weeks was evaluated separately and in combination on 633 recruited patients. AUC for sPTB prediction before weeks 37 and 34 respectively were as follows: 55.5% and 65.3% for CL, 63.4% and 66.3% for texture analysis, 67.5% and 76.7% when combined. The new test improved detection rates of CL at similar low FPR. Combining the two increased detection rate compared to CL alone from 13.0 to 30.4% for sPTB < 37 and from 14.3 to 42.9% sPTB < 34. Texture analysis of cervical ultrasound improved sPTB detection rate compared to cervical length for similar FPR, and the two combined together increased significantly prediction performance. This results should be confirmed in larger cohorts.

The Predictive Value of the Cervical Consistency Index to Predict Spontaneous Preterm Birth in Asymptomatic Twin Pregnancies at the Second-Trimester Ultrasound Scan: A Prospective Cohort Study

Novel transvaginal ultrasound (TVU) markers have been proposed to improve spontaneous preterm birth (sPTB) prediction. Preliminary results of the cervical consistency index (CCI), uterocervical angle (UCA), and cervical texture (CTx) have been promising in singletons. However, in twin pregnancies, the results have been inconsistent. In this prospective cohort study of asymptomatic twin pregnancies assessed between 18⁺⁰–22⁺⁰ weeks, we evaluated TVU derived cervical length (CL), CCI, UCA, and the CTx to predict sPTB < 34⁺⁰ weeks. All iatrogenic PTB were excluded. In the final cohort of 63 pregnancies, the sPTB rate < 34⁺⁰ was 16.3%. The CCI, UCA, and CTx, including the CL was significantly different in the sPTB < 34⁺⁰ weeks group. The best area under the receiver operating characteristic curve (AUC) for sPTB < 34⁺⁰ weeks was achieved by the CCI 0.82 (95%CI, 0.72–0.93), followed by the UCA with AUC 0.72 (95%CI, 0.57–0.87). A logistic regression model incorporating parity, chorionicity, CCI, and UCA resulted in an AUC of 0.91 with a sensitivity of 55.3% and specificity of 88.1% for predicting sPTB < 34⁺⁰. The CCI performed better than other TVU markers to predict sPTB < 34⁺⁰ in twin gestations, and the best diagnostic accuracy was achieved by a combination of parity, chorionicity, CCI, and UCA.

Intra- and interobserver reproducibility of second trimester ultrasound cervical length measurement in a general population

Objectives: To evaluate the reproducibility of ultrasound cervical length (CL) measurement at the second trimester.Methods: A set of 565 cervical ultrasound images were collected at 19 + 0-24 + 6 weeks' gestation. Two senior maternal-fetal specialists measured CL in each image on three occasions 2 weeks apart. In the interval between the first and following two measures, the clinicians reviewed 20 images together to agree on the criteria for measurement. Measurements were analyzed for intra- and inter-observer disagreement. The robustness of patient classification when CL measure was used with different cutoff thresholds was analyzed.Results: Average intra-observer deviation was 2.8 mm for clinician 1 and 3.7 mm for clinician 2. Inter-observer deviation among the two clinicians was 5.2 and 3.2 mm before and after reviewing measurement criteria together. When cutoffs were used to classify as "short" cervix, disagreement ranged from 22 to 70% depending on operator and threshold used.Conclusion: Ultrasound CL measurements by experts showed moderate intra- and inter-observer reproducibility. The use of specific cutoffs to classify patients as high or low risk resulted in wide disagreements. The results stress the importance of training and quality assessments on considering universal screening application of CL measurement.

Quantitative Ultrasound Image Analysis of Axillary Lymph Nodes to Diagnose Metastatic Involvement in Breast Cancer

This study aimed to assess the potential of state-of-the-art ultrasound analysis techniques to non-invasively diagnose axillary lymph nodes involvement in breast cancer. After exclusion criteria, 105 patients were selected from two different hospitals. The 118 lymph node ultrasound images taken from these patients were divided into 53 cases and 65 controls, which made up the study series. The clinical outcome of each node was verified by ultrasound-guided fine needle aspiration, core needle biopsy or surgical biopsy. The achieved accuracy of the proposed method was 86.4%, with 84.9% sensitivity and 87.7% specificity. When tested on breast cancer patients only, the proposed method improved the accuracy of the sonographic assessment of axillary lymph nodes performed by expert radiologists by 9% (87.0% vs 77.9%). In conclusion, the results demonstrate the potential of ultrasound image analysis to detect the microstructural and compositional changes that occur in lymph nodes because of metastatic involvement.

Automated measurement of mouse social behaviors using depth sensing, video tracking, and machine learning

A lack of automated, quantitative, and accurate assessment of social behaviors in mammalian animal models has limited progress toward understanding mechanisms underlying social interactions and their disorders such as autism. Here we present a new integrated hardware and software system that combines video tracking, depth sensing, and machine learning for automatic detection and quantification of social behaviors involving close and dynamic interactions between two mice of different coat colors in their home cage. We designed a hardware setup that integrates traditional video cameras with a depth camera, developed computer vision tools to extract the body "pose" of individual animals in a social context, and used a supervised learning algorithm to classify several well-described social behaviors. We validated the robustness of the automated classifiers in various experimental settings and used them to examine how genetic background, such as that of Black and Tan Brachyury (BTBR) mice (a previously reported autism model), influences social behavior. Our integrated approach allows for rapid, automated measurement of social behaviors across diverse experimental designs and also affords the ability to develop new, objective behavioral metrics.

Mapping wide row crops with video sequences acquired from a tractor moving at treatment speed

This paper presents a mapping method for wide row crop fields. The resulting map shows the crop rows and weeds present in the inter-row spacing. Because field videos are acquired with a camera mounted on top of an agricultural vehicle, a method for image sequence stabilization was needed and consequently designed and developed. The proposed stabilization method uses the centers of some crop rows in the image sequence as features to be tracked, which compensates for the lateral movement (sway) of the camera and leaves the pitch unchanged. A region of interest is selected using the tracked features, and an inverse perspective technique transforms the selected region into a bird’s-eye view that is centered on the image and that enables map generation. The algorithm developed has been tested on several video sequences of different fields recorded at different times and under different lighting conditions, with good initial results. Indeed, lateral displacements of up to 66% of the inter-row spacing were suppressed through the stabilization process, and crop rows in the resulting maps appear straight.

An image segmentation based on a genetic algorithm for determining soil coverage by crop residues

Determination of the soil coverage by crop residues after ploughing is a fundamental element of Conservation Agriculture. This paper presents the application of genetic algorithms employed during the fine tuning of the segmentation process of a digital image with the aim of automatically quantifying the residue coverage. In other words, the objective is to achieve a segmentation that would permit the discrimination of the texture of the residue so that the output of the segmentation process is a binary image in which residue zones are isolated from the rest. The RGB images used come from a sample of images in which sections of terrain were photographed with a conventional camera positioned in zenith orientation atop a tripod. The images were taken outdoors under uncontrolled lighting conditions. Up to 92% similarity was achieved between the images obtained by the segmentation process proposed in this paper and the templates made by an elaborate manual tracing process. In addition to the proposed segmentation procedure and the fine tuning procedure that was developed, a global quantification of the soil coverage by residues for the sampled area was achieved that differed by only 0.85% from the quantification obtained using template images. Moreover, the proposed method does not depend on the type of residue present in the image. The study was conducted at the experimental farm "El Encín" in Alcalá de Henares (Madrid, Spain).