Patient's airway monitoring during cardiopulmonary resuscitation using deep networks.

Cardiopulmonary resuscitation (CPR) is a crucial life-saving technique commonly administered to individuals experiencing cardiac arrest. Among the important aspects of CPR is ensuring the correct airway position of the patient, which is typically monitored by human tutors or supervisors. This study aims to utilize deep transfer learning for the detection of the patient's correct and incorrect airway position during cardiopulmonary resuscitation. To address the challenge of identifying the airway position, we curated a dataset consisting of 198 recorded video sequences, each lasting 6-8 s, showcasing both correct and incorrect airway positions during mouth-to-mouth breathing and breathing with an Ambu Bag. We employed six cutting-edge deep networks, namely DarkNet19, EfficientNetB0, GoogleNet, MobileNet-v2, ResNet50, and NasnetMobile. These networks were initially pre-trained on computer vision data and subsequently fine-tuned using the CPR dataset. The validation of the fine-tuned networks in detecting the patient's correct airway position during mouth-to-mouth breathing achieved impressive results, with the best sensitivity (98.8 %), specificity (100 %), and F-measure (97.2 %). Similarly, the detection of the patient's correct airway position during breathing with an Ambu Bag exhibited excellent performance, with the best sensitivity (100 %), specificity (99.8 %), and F-measure (99.7 %).

Learning-to-augment incorporated noise-robust deep CNN for detection of COVID-19 in noisy X-ray images.

Deep convolutional neural networks (CNNs) are used for the detection of COVID-19 in X-ray images. The detection performance of deep CNNs may be reduced by noisy X-ray images. To improve the robustness of a deep CNN against impulse noise, we propose a novel CNN approach using adaptive convolution, with the aim to ameliorate COVID-19 detection in noisy X-ray images without requiring any preprocessing for noise removal. This approach includes an impulse noise-map layer, an adaptive resizing layer, and an adaptive convolution layer to the conventional CNN framework. We also used a learning-to-augment strategy using noisy X-ray images to improve the generalization of a deep CNN. We have collected a dataset of 2093 chest X-ray images including COVID-19 (452 images), non-COVID pneumonia (621 images), and healthy ones (1020 images). The architecture of pre-trained networks such as SqueezeNet, GoogleNet, MobileNetv2, ResNet18, ResNet50, ShuffleNet, and EfficientNetb0 has been modified to increase their robustness to impulse noise. Validation on the noisy X-ray images using the proposed noise-robust layers and learning-to-augment strategy-incorporated ResNet50 showed 2% better classification accuracy compared with state-of-the-art method.

Active deep learning from a noisy teacher for semi-supervised 3D image segmentation: Application to COVID-19 pneumonia infection in CT.

Supervised deep learning has become a standard approach to solving medical image segmentation tasks. However, serious difficulties in attaining pixel-level annotations for sufficiently large volumetric datasets in real-life applications have highlighted the critical need for alternative approaches, such as semi-supervised learning, where model training can leverage small expert-annotated datasets to enable learning from much larger datasets without laborious annotation. Most of the semi-supervised approaches combine expert annotations and machine-generated annotations with equal weights within deep model training, despite the latter annotations being relatively unreliable and likely to affect model optimization negatively. To overcome this, we propose an active learning approach that uses an example re-weighting strategy, where machine-annotated samples are weighted (i) based on the similarity of their gradient directions of descent to those of expert-annotated data, and (ii) based on the gradient magnitude of the last layer of the deep model. Specifically, we present an active learning strategy with a query function that enables the selection of reliable and more informative samples from machine-annotated batch data generated by a noisy teacher. When validated on clinical COVID-19 CT benchmark data, our method improved the performance of pneumonia infection segmentation compared to the state of the art.

Learnable image histograms-based deep radiomics for renal cell carcinoma grading and staging.

Fuhrman cancer grading and tumor-node-metastasis (TNM) cancer staging systems are typically used by clinicians in the treatment planning of renal cell carcinoma (RCC), a common cancer in men and women worldwide. Pathologists typically use percutaneous renal biopsy for RCC grading, while staging is performed by volumetric medical image analysis before renal surgery. Recent studies suggest that clinicians can effectively perform these classification tasks non-invasively by analyzing image texture features of RCC from computed tomography (CT) data. However, image feature identification for RCC grading and staging often relies on laborious manual processes, which is error prone and time-intensive. To address this challenge, this paper proposes a learnable image histogram in the deep neural network framework that can learn task-specific image histograms with variable bin centers and widths. The proposed approach enables learning statistical context features from raw medical data, which cannot be performed by a conventional convolutional neural network (CNN). The linear basis function of our learnable image histogram is piece-wise differentiable, enabling back-propagating errors to update the variable bin centers and widths during training. This novel approach can segregate the CT textures of an RCC in different intensity spectra, which enables efficient Fuhrman low (I/II) and high (III/IV) grading as well as RCC low (I/II) and high (III/IV) staging. The proposed method is validated on a clinical CT dataset of 159 patients from The Cancer Imaging Archive (TCIA) database, and it demonstrates 80% and 83% accuracy in RCC grading and staging, respectively.

Cascaded Regression Neural Nets for Kidney Localization and Segmentation-free Volume Estimation.

Kidney volume is an essential biomarker for a number of kidney disease diagnoses, for example, chronic kidney disease. Existing total kidney volume estimation methods often rely on an intermediate kidney segmentation step. On the other hand, automatic kidney localization in volumetric medical images is a critical step that often precedes subsequent data processing and analysis. Most current approaches perform kidney localization via an intermediate classification or regression step. This paper proposes an integrated deep learning approach for (i) kidney localization in computed tomography scans and (ii) segmentation-free renal volume estimation. Our localization method uses a selection-convolutional neural network that approximates the kidney inferior-superior span along the axial direction. Cross-sectional (2D) slices from the estimated span are subsequently used in a combined sagittal-axial Mask-RCNN that detects the organ bounding boxes on the axial and sagittal slices, the combination of which produces a final 3D organ bounding box. Furthermore, we use a fully convolutional network to estimate the kidney volume that skips the segmentation procedure. We also present a mathematical expression to approximate the 'volume error' metric from the 'Sørensen-Dice coefficient.' We accessed 100 patients' CT scans from the Vancouver General Hospital records and obtained 210 patients' CT scans from the 2019 Kidney Tumor Segmentation Challenge database to validate our method. Our method produces a kidney boundary wall localization error of ~2.4mm and a mean volume estimation error of ~5%.

Learning-to-augment strategy using noisy and denoised data: Improving generalizability of deep CNN for the detection of COVID-19 in X-ray images.

Chest X-ray images are used in deep convolutional neural networks for the detection of COVID-19, the greatest human challenge of the 21st century. Robustness to noise and improvement of generalization are the major challenges in designing these networks. In this paper, we introduce a strategy for data augmentation using the determination of the type and value of noise density to improve the robustness and generalization of deep CNNs for COVID-19 detection. Firstly, we present a learning-to-augment approach that generates new noisy variants of the original image data with optimized noise density. We apply a Bayesian optimization technique to control and choose the optimal noise type and its parameters. Secondly, we propose a novel data augmentation strategy, based on denoised X-ray images, that uses the distance between denoised and original pixels to generate new data. We develop an autoencoder model to create new data using denoised images corrupted by the Gaussian and impulse noise. A database of chest X-ray images, containing COVID-19 positive, healthy, and non-COVID pneumonia cases, is used to fine-tune the pre-trained networks (AlexNet, ShuffleNet, ResNet18, and GoogleNet). The proposed method performs better results compared to the state-of-the-art learning to augment strategies in terms of sensitivity (0.808), specificity (0.915), and F-Measure (0.737). The source code of the proposed method is available at https://github.com/mohamadmomeny/Learning-to-augment-strategy.

Strain-Initialized Robust Bone Surface Detection in 3-D Ultrasound.

Three-dimensional ultrasound has been increasingly considered as a safe radiation-free alternative to radiation-based fluoroscopic imaging for surgical guidance during computer-assisted orthopedic interventions, but because ultrasound images contain significant artifacts, it is challenging to automatically extract bone surfaces from these images. We propose an effective way to extract 3-D bone surfaces using a surface growing approach that is seeded from 2-D bone contours. The initial 2-D bone contours are estimated from a combination of ultrasound strain images and envelope power images. Novel features of the proposed method include: (i) improvement of a previously reported 2-D strain imaging-based bone segmentation method by incorporation of a depth-dependent cumulative power of the envelope into the elastographic data; (ii) incorporation of an echo decorrelation measure-based weight to fuse the strain and envelope maps; (iii) use of local statistics of the bone surface candidate points to detect the presence of any bone discontinuity; and (iv) an extension of our 2-D bone contour into a 3-D bone surface by use of an effective surface growing approach. Our new method produced average improvements in the mean absolute error of 18% and 23%, respectively, on 2-D and 3-D experimental phantom data, compared with those of two state-of-the-art bone segmentation methods. Validation on 2-D and 3-D clinical in vivo data also reveals, respectively, an average improvement in the mean absolute fitting error of 55% and an 18-fold improvement in the computation time.

Robust bone detection in ultrasound using combined strain imaging and envelope signal power detection.

Bone localization in ultrasound (US) remains challenging despite encouraging advances. Current methods, e.g. local image phase-based feature analysis, showed promising results but remain reliant on delicate parameter selection processes and prone to errors at confounding soft tissue interfaces of similar appearance to bone interfaces. We propose a different approach combining US strain imaging and envelope power detection at each radio-frequency (RF) sample. After initial estimation of strain and envelope power maps, we modify their dynamic ranges into a modified strain map (MSM) and a modified envelope map (MEM) that we subsequently fuse into a single combined map that we show corresponds robustly to actual bone boundaries. Our quantitative results demonstrate a marked reduction in false positive responses at soft tissue interfaces and an increase in bone delineation accuracy. Comparisons to the state-of-the-art on a finite-element-modelling (FEM) phantom and fiducial-based experimental phantom show an average improvement in mean absolute error (MAE) between actual and estimated bone boundaries of 32% and 14%, respectively. We also demonstrate an average reduction in false bone responses of 87% and 56%, respectively. Finally, we qualitatively validate on clinical in vivo data of the human radius and ulna bones, and demonstrate similar improvements to those observed on phantoms.

Lesion edge preserved direct average strain estimation for ultrasound elasticity imaging.

Elasticity imaging techniques with built-in or regularization-based smoothing feature for ensuring strain continuity are not intelligent enough to prevent distortion or lesion edge blurring while smoothing. This paper proposes a novel approach with built-in lesion edge preservation technique for high quality direct average strain imaging. An edge detection scheme, typically used in diffusion filtering is modified here for lesion edge detection. Based on the extracted edge information, lesion edges are preserved by modifying the strain determining cost function in the direct-average-strain-estimation (DASE) method. The proposed algorithm demonstrates approximately 3.42-4.25 dB improvement in terms of edge-mean-square-error (EMSE) than the other reported regularized or average strain estimation techniques in finite-element-modeling (FEM) simulation with almost no sacrifice in elastographic-signal-to-noise-ratio (SNRe) and elastographic-contrast-to-noise-ratio (CNRe) metrics. The efficacy of the proposed algorithm is also tested for the experimental phantom data and in vivo breast data. The results reveal that the proposed method can generate a high quality strain image delineating the lesion edge more clearly than the other reported strain estimation techniques that have been designed to ensure strain continuity. The computational cost, however, is little higher for the proposed method than the simpler DASE and considerably higher than that of the 2D analytic minimization (AM2D) method.

Using nearest neighbors for accurate estimation of ultrasonic attenuation in the spectral domain.

Attenuation is a key diagnostic parameter of tissue pathology change and thus may play a vital role in the quantitative discrimination of malignant and benign tumors in soft tissue. In this paper, two novel techniques are proposed for estimating the average ultrasonic attenuation in soft tissue using the spectral domain weighted nearest neighbor method. Because the attenuation coefficient of soft tissues can be considered to be a continuous function in a small neighborhood, we directly estimate an average value of it from the slope of the regression line fitted to the 1) modified average midband fit value and 2) the average center frequency shift along the depth. To calculate the average midband fit value, an average regression line computed from the exponentially weighted short-time Fourier transform (STFT) of the neighboring 1-D signal blocks, in the axial and lateral directions, is fitted over the usable bandwidth of the normalized power spectrum. The average center frequency downshift is computed from the maximization of a cost function defined from the normalized spectral cross-correlation (NSCC) of exponentially weighted nearest neighbors in both directions. Different from the large spatial signal-block-based spectral stability approach, a costfunction- based approach incorporating NSCC functions of neighboring 1-D signal blocks is introduced. This paves the way for using comparatively smaller spatial area along the lateral direction, a necessity for producing more realistic attenuation estimates for heterogeneous tissue. For accurate estimation of the attenuation coefficient, we also adopt a reference-phantombased diffraction-correction technique for both methods. The proposed attenuation estimation algorithm demonstrates better performance than other reported techniques in the tissue-mimicking phantom and the in vivo breast data analysis.

Robust strain-estimation algorithm using combined radiofrequency and envelope cross-correlation with diffusion filtering.

In ultrasound elastography, the strain in compressed tissue due to external deformation is estimated and is smaller in harder than softer tissue. With increased stress, the nonaxial motions of tissue elements increase and result in noisier strain images. At high strain, the envelope of the rf signal exhibits robustness to signal decorrelation. However, the precision of strain estimates using envelope signals is much worse compared to that using the rf signals. In this paper, we propose a novel approach for robust strain estimation by combining weighted rf cross-correlation and envelope cross-correlation functions. An applied strain-dependent piecewise-linear-weight is used for this purpose. In addition, we introduce nonlinear diffusion filtering to further enhance the resulting strain image. The results of our algorithm are demonstrated for up to 10% applied strain using a finite-element modelling (FEM) simulation phantom. It reveals that the elastographic signal-to-noise ratio (SNRe) and the elastographic contrast-to-noise ratio (CNRe) of the strain images can be improved more significantly than with other algorithms used in this paper. In addition, comparative results in terms of the mean structural similarity (MSSIM) using in vivo breast data show that the strain image quality can be improved noticeably by the proposed method than with the techniques employed in this work.

Direct and gradient-based average strain estimation by using weighted nearest neighbor cross-correlation peaks.

In this paper, two novel approaches, gradient-based and direct strain estimation techniques, are proposed for high-quality average strain imaging incorporating a cost function maximization. Stiffness typically is a continuous function. Consequently, stiffness of proximal tissues is very close to that of the tissue corresponding to a given data window. Hence, a cost function is defined from exponentially weighted neighboring pre- and post-compression RF echo normalized cross-correlation peaks in the lateral (for displacement estimation) or in both the axial and the lateral (for direct strain estimation) directions. This enforces a controlled continuity in displacement/strain and average displacement/strain is calculated from the corresponding maximized cost function. Axial stress causes lateral shift in the tissue. Therefore, a 1-D post-compression echo segment is selected by incorporating Poisson's ratio. Two stretching factors are considered simultaneously in gradient-based strain estimation that allow imaging the lesions properly. The proposed time-domain gradient-based and direct-strain-estimation-based algorithms demonstrate significantly better performance in terms of elastographic signal-to-noise ratio (SNRe), elastographic contrast-to-noise ratio (CNRe), peak signal-to-noise ratio (PSNR), and mean structural similarity (MSSIM) than the other reported time-domain gradient-based and direct-strain-estimation techniques in finite element modeling (FEM) simulation and phantom experiments. For example, in FEM simulation, it has been found that the proposed direct strain estimation method can improve up to approximately 2.49 to 8.71, 2.2 to 6.63, 1.5 to 5, and 1.59 to 2.45 dB in the SNRe, CNRe, PSNR, and MSSIM compared with the traditional direct strain estimation method, respectively, and the proposed gradient-based algorithm demonstrates 2.99 to 16.26, 18.74 to 23.88, 3 to 9.5, and 0.6 to 5.36 dB improvement in the SNRe, CNRe, PSNR, and MSSIM, respectively, compared with a recently reported time-domain gradient-based technique. The range of improvement as noted above is for low to high applied strains. In addition, the comparative results using the in vivo breast data (including malignant or benign masses) also show that the lesion size is better defined by the proposed gradient-based average strain estimation technique.