Acoustic shadows are common artifacts in medical ultrasound imaging. The shadows are caused by objects that reflect ultrasound such as bones, and they are shown as dark areas in ultrasound images. Detecting such shadows is crucial for assessing the quality of images. This will be a pre-processing for further image processing or recognition aiming computer-aided diagnosis. In this paper, we propose an auto-encoding structure that estimates the shadowed areas and their intensities. The model once splits an input image into an estimated shadow image and an estimated shadow-free image through its encoder and decoder. Then, it combines them to reconstruct the input. By generating plausible synthetic shadows based on relatively coarse domain-specific knowledge on ultrasound images, we can train the model using unlabeled data. If pixel-level labels of the shadows are available, we also utilize them in a semi-supervised fashion. By experiments on ultrasound images for fetal heart diagnosis, we show that our method achieved 0.720 in the DICE score and outperformed conventional image processing methods and a segmentation method based on deep neural networks. The capability of the proposed method on estimating the intensities of shadows and the shadow-free images is also indicated through the experiments.

/pdf/shadow-estimation-for-ultrasound-images-using-auto-encoding-56pkpej8ho.pdf

Shadow Estimation for Ultrasound Images Using Auto-Encoding Structures and Synthetic Shadows

/pdf/a-review-on-deep-learning-algorithms-for-fetal-ultrasound-ejqth3ya.pdf

A Review on Deep-Learning Algorithms for Fetal Ultrasound-Image Analysis

Ultrasound is one of the most ubiquitous imaging modalities in clinical practice It is cheap, does not require ionizing radiation and can be performed at the bedside, making it the most commonly utilized imaging technique in pregnancy Despite these advantages, it does have some drawbacks such as relatively low imaging quality, low contrast, and high variability With these constraints, automating the interpretation of ultrasound images is challenging However, successful automated identification of structures within 3D ultrasound volumes has the potential to revolutionize clinical practice For example, a small placental volume in the first trimester has been shown to be correlated to adverse outcome later in pregnancy If the placenta could be segmented reliably and automatically from a static 3D ultrasound volume, it would facilitate the use of its estimated volume, and other morphological metrics, as part of a screening test for increased risk of pregnancy complications potentially improving clinical outcomes Recently, deep learning has emerged, achieving state-of-the-art performance in various research fields, notably medical image analysis involving classification, segmentation, object detection, and tracking tasks Due to its increased performance with large datasets, it has gained great interest in medical imaging applications In this review, we present an overview of deep learning methods applied to ultrasound in pregnancy, introducing their architectures and analyzing their strategies We then present some common problems and provide some perspectives into potential future research

Deep Learning strategies for Ultrasound in Pregnancy.

Placenta previa describes the abnormal implantation of the placenta in the uterus of pregnant women, entirely or partially obstructing the internal cervical os. For clinical significance, it is suggested to assess the final placental location in the third trimester. In rural areas of developing countries, the absence of radiologists and sophisticated ultrasound systems make difficult the diagnosis of this condition. Ten pregnant women in their third trimester (age range of 22–43 years) were enrolled in this study. A volume sweep imaging obstetric protocol based on eight free-hand sweeping (three transverse and five longitudinal throughout the maternal abdominal area) was used. The acquisition was performed using a portable ultrasound scanner by a non-specialist. The dataset consisted of eight videos per volunteer and one hundred frames per video on average. Placental segmentation was carried out using the U-Net deep learning architecture and the leave one out cross-validation approach. The output masks combined with the knowledge of the acquisition protocol allowed to assess the spatial location likelihood of the placenta in the uterus, which is represented by a heat map. The method showed a sensitivity of 75% and a specificity of 92 %. Albeit the need for further validation, this approach may automatically warn of possible risks in childbirth, giving the chance to notify the specialist.

Automatic ultrasound assessment of placenta previa during the third trimester for rural areas

/pdf/placenta-segmentation-in-ultrasound-imaging-addressing-1f1iiu00.pdf

Placenta segmentation in ultrasound imaging: Addressing sources of uncertainty and limited field-of-view

Placental assessment through routine obstetrical ultrasound is often limited to documenting its location and ruling out placenta previa. However, many obstetrical complications originate from abnormal focal or global placental development. Technical difficulties in assessing the placenta as well as a lack of established objective criteria to classify echotexture are barriers to diagnosis of pathology by ultrasound imaging. As a first step towards the development of a computer aided placental assessment tool, we developed a fully automated method for placental segmentation using a convolutional neural network. The network contains a novel layer weighted by automated acoustic shadow detection to recognize artifacts specific to ultrasound. In order to develop a detection algorithm usable in different imaging scenarios, we acquired a dataset containing 1364 fetal ultrasound images from 247 patients acquired over 47 months was taken with different machines, operators, and at a range of gestational ages. Mean Dice coefficients for automated segmentation on the full dataset with and without the acoustic shadow detection layer were 0.92±0.04 and 0.91±0.03 when comparing to manual segmentation. Mean Dice coefficients on the subset of images containing acoustic shadows with and without acoustic shadow detection were 0.87±0.04 and 0.75±0.05. The method requires no user input to tune the detection. The automated placenta segmentation method can serve as a preprocessing step for further image analysis in artificial intelligence methods requiring large scale data processing of placental images.

Automated Placenta Segmentation with a Convolutional Neural Network Weighted by Acoustic Shadow Detection

Diseases related to the placenta, such as preeclampsia (PE) and fetal growth restriction (FGR), are major causes of mortality and morbidity. Diagnostic criteria of such diseases are defined by biomarkers, such as proteinuria, that appear in advanced gestational age. As placentally-mediated disease is often clinically unrecognized until later stages, accurate early diagnosis is required to allow earlier intervention, which is particularly challenging in low-resource areas without subspecialty clinicians. Proposed attempts at early diagnosis involve a combination of subjective and objective ultrasound placental assessments which have limited accuracy and high interobserver variability. Machine learning, particularly with convolutional neural networks, have shown potential in analyzing complex textural features in ultrasound imaging that may be predictive of disease. We propose a model utilizing a two-stage convolutional neural network pipeline to classify the presence of placental disease. The pipeline involves a segmentation stage to extract the placenta followed by a classification stage. We evaluated the pipeline on retrospectively collected placenta ultrasound scans and diagnostic outcomes of 321 patients taken by 18 sonographers and 3 ultrasound machines. Compared to existing clinical algorithms and neural networks, our classifier achieved significantly higher accuracy of 0.81 ± 0.02 (p < 0.05). Class activation maps were generated to identify potential abnormal regions of interest in placenta tissue. This study provides support that automated image analysis of ultrasound texture may assist physicians in early identification of placental disease, with potential benefits to low-resource environments.

Ryan Yan

Papers

Automated Placenta Segmentation with a Convolutional Neural Network Weighted by Acoustic Shadow Detection

Automatic Placenta Abnormality Detection Using Convolutional Neural Networks on Ultrasound Texture