In recent years, there has been a growing trend in accelerating computationally complex nonreal-time beamforming algorithms in ultrasound imaging using deep learning models. However, due to the large size and complexity, these state-of-the-art deep learning techniques pose significant challenges when deploying on resource-constrained edge devices. In this work, we propose a novel capsule network-based beamformer called CapsBeam, designed to operate on raw radio frequency data and provide an envelope of beamformed data through nonsteered plane-wave insonification. In experiments on in vivo data, CapsBeam reduced artifacts compared to the standard Delay-and-Sum (DAS) beamforming. For in vitro data, CapsBeam demonstrated a 32.31% increase in contrast, along with gains of 16.54% and 6.7% in axial and lateral resolution compared to the DAS. Similarly, in silico data showed a 26% enhancement in contrast, along with improvements of 13.6% and 21.5% in axial and lateral resolution, respectively, compared to the DAS. To reduce the parameter redundancy and enhance the computational efficiency, we pruned the model using our multilayer look-ahead kernel pruning (LAKP-ML) methodology, achieving a compression ratio of 85% without affecting the image quality. Additionally, the hardware complexity of the proposed model is reduced by applying quantization, simplification of nonlinear operations, and parallelizing operations. Finally, we proposed a specialized accelerator architecture for the pruned and optimized CapsBeam model, implemented on a Xilinx ZU7EV FPGA. The proposed accelerator achieved a throughput of 30 GOPS for the convolution operation and 17.4 GOPS for the dynamic routing operation.

Accelerating compute intensive non-real-time beam-forming algorithms in ultrasound imaging using deep learning architectures has been gaining momentum in the recent past. Nonetheless, the complexity of the state-of-the-art deep learning techniques poses challenges for deployment on resource-constrained edge devices. In this work, we propose a novel vision transformer based tiny beamformer (Tiny-VBF), which works on the raw radio-frequency channel data acquired through single-angle plane wave insonification. The output of our Tiny-VBF provides fast envelope detection requiring very low frame rate, i.e. 0.34 GOPs/Frame for a frame size of 368 × 128 in comparison to the state-of-the-art deep learning models. It also exhibited an 8% increase in contrast and gains of 5% and 33% in axial and lateral resolution respectively when compared to Tiny-CNN on in-vitro dataset. Additionally, our model showed a 4.2% increase in contrast and gains of 4% and 20% in axial and lateral resolution respectively when compared against conventional Delay-and-Sum (DAS) beamformer. We further propose an accelerator architecture and implement our Tiny-VBF model on a Zynq UltraScale+ MPSoC ZCU104 FPGA using a hybrid quantization scheme with 50% less resource consumption compared to the floating-point implementation, while preserving the image quality.

Field programmable gate array (FPGA) implementation of delay and sum beamforming (DASB) for ultrasound (US) imaging systems has been proposed in the literature. However, the homogeneous diffuse assumption in estimating receive sensor weights in DASB is violated for specular reflectors like bones or needles in guided interventions which generate highly directive reflections challenging their visualization in the US images. To address this, apodization coefficients need to be dynamically determined according to reflection directivity. This work introduces a novel approach called raster apodization design (RAD) to dynamically adjust apodization coefficients at the pixel level on hardware, taking into account the directivity of reflections. RAD allows for the simultaneous estimation of apodization coefficients for all pixels at a fixed depth, resembling a raster scan while reusing a single apodization window. The design is implemented on a Xilinx XCZ7010clg400-1 FPGA and compares resource and power estimates at pixel level and for a region of interest. A comparison of the performance of DASB with hardware and software RAD coefficients is also illustrated to emphasize the accuracy and efficiency of the proposed design.

The COVID-19 pandemic has highlighted the need for a tool to speed up triage in ultrasound scans and provide clinicians with fast access to relevant information. To this end, we propose a new unsupervised reinforcement learning (RL) framework with novel rewards to facilitate unsupervised learning by avoiding tedious and impractical manual labelling for summarizing ultrasound videos. The proposed framework is capable of delivering video summaries with classification labels and segmentations of key landmarks which enhances its utility as a triage tool in the emergency department (ED) and for use in telemedicine. Using an attention ensemble of encoders, the high dimensional image is projected into a low dimensional latent space in terms of: a) reduced distance with a normal or abnormal class (classifier encoder), b) following a topology of landmarks (segmentation encoder), and c) the distance or topology agnostic latent representation (autoencoders). The summarization network is implemented using a bi-directional long short term memory (Bi-LSTM) which utilizes the latent space representation from the encoder. Validation is performed on lung ultrasound (LUS), that typically represent potential use cases in telemedicine and ED triage acquired from different medical centers across geographies (India and Spain). The proposed approach trained and tested on 126 LUS videos showed high agreement with the ground truth with an average precision of over 80% and average$\boldsymbol{F}_{{1}}$score of well over$\boldsymbol{44 \pm 1.7 \%}$. The approach resulted in an average reduction in storage space of 77% which can ease bandwidth and storage requirements in telemedicine.

With the recent outbreak of COVID-19, ultrasound is fast becoming an inevitable diagnostic tool for regular and continuous monitoring of the lung. However, lung ultrasound (LUS) is unique in the perspective that, the artefacts created by acoustic wave propagation is aiding clinicians in diagnosis. In this work, a novel approach is presented to extract acoustic wave propagation driven features such as acoustic shadows, local phase-based feature symmetry, and integrated backscattering to automatically detect the pleura and to aid a pretrained neural network to classify the severity of lung infection based on the region below pleura. A detailed analysis of the proposed approach on LUS images over the infection to full recovery period of ten confirmed COVID-19 subjects across 400 videos shows an average five-fold cross-validation accuracy, sensitivity, and specificity of 97%, 92%, and 98% respectively over randomly selected 5000 frames. The results and analysis show that, when the input dataset is limited and diverse as in the case of COVID-19 pandemic, an aided effort of combining acoustic propagation-based features along with the gray scale images, as proposed in this work, improves the performance of the neural network significantly even when tested against a completely new data acquisition.

In ultrasound imaging, a non-linear beamforming algorithm called filtered delay multiply and sum (F-DMAS) has been demonstrated to provide improved contrast and resolution compared to the commonly employed delay and sum (DAS) algorithm. In F-DMAS, the delay compensated radio-frequency narrowband signals from the transducer channels is pairwise multiplied to generate baseband and higher-order harmonic components in the output spectrum. The final image is reconstructed by filtering the second harmonic component to provide better contrast and resolution. However, other generated harmonic components, which are not utilized in the standard F-DMAS could be employed for improved contrast and resolution. In this work, a modification to standard F-DMAS is proposed where the different frequency bands generated through pairwise multiplications are selectively combined through additive or difference compounding techniques to form the final image. The results show that the proposed approach outperforms the standard F-DMAS in terms of contrast and resolution.

Musculoskeletal (MSK) ultrasound imaging aims to provide pictures of tissues and bones such as muscles, tendons, ligaments, joints and soft tissues throughout the body. One of the major landmarks in MSK ultrasound are the bones, and segmentation of bone surface has numerous applications in computer-aided orthopedic diagnosis. In this work, a novel method of bone aware image enhancement of MSK ultrasound images is presented. A combination of fundamental and harmonic US images is used for bone segmentation. The method for bone segmentation takes into account the acoustic characteristics of the intensity of bones used for computing their acoustic shadows, local phase-based features such as local energy, local phase, and feature symmetry based on a reported work in literature. It is combined with integrated backscattering of the bone to provide a probability map of the bone. Bone location in probability map was found based on the centroid of the intensity distribution. Further, image enhancement of the extracted region of interest based on the bone for distinctive visualization of the muscular and tendon region above the bone structure is presented. The image enhancement techniques employed are gamma correction, histogram equalization, adaptive histogram equalization and an improved frequency based super-resolution of ultrasound images.