Sajjad Afrakhteh

Class imbalance is a significant challenge in medical image analysis, particularly in lung ultrasound (LUS), where severe patterns are often underrepresented. Traditional oversampling techniques, which simply duplicate original data, have limited effectiveness in addressing this issue. To overcome these limitations, this study introduces a novel supervised autoencoder generative adversarial network (SA-GAN) for data augmentation, leveraging advanced generative AI to create high-quality synthetic samples for minority classes. Additionally, the traditional data augmentation technique is employed for comparison. The SA-GAN incorporates an autoencoder to develop a conditional latent space, effectively addressing weight clipping issues and ensuring higher-quality synthetic data. The generated samples are evaluated using similarity metrics and expert analysis to validate their utility. Furthermore, state-of-the-art neural networks are employed for multi-class classification, and their performance is compared when trained with GAN-based augmentation versus traditional data augmentation techniques. These contributions enhance the robustness and reliability of AI models in mitigating class imbalance in LUS analysis.

Ultrasound localization microscopy (ULM) has become a potent technique for microvascular imaging using ultrasound waves. However, one major challenge is the high frame rate and lengthy acquisition time needed to produce super-resolved (SR) images. To overcome this, our goal is to relax the frame rate and shorten this acquisition time while preserving SR image quality, thereby enhancing ULM’s clinical applicability. To this end, we propose two distinct strategies: first, we suggest acquiring the data at lower frame rate followed by applying the reconstruction technique to compensate the lost information due to the low frame rate imaging. Secondly, to tackle the prolonged acquisition time, we propose compressing acquisition time by a compression ratio (CR), which can degrade SR image quality due to reduced temporal information. To mitigate this, we temporally upsample the in-phase-quadrature (IQ) data by a factor equal to the CR after the compressed acquisition. Additionally, we introduce a novel bi-directional (2x2D) interpolation using radial basis function (RBF)-based reconstruction to estimate unknown values in the 3D IQ data (x-z-t), thereby enhancing temporal resolution. The rationale behind using 2x2D interpolation is its ability to integrate spatiotemporal information from two orthogonal x-t and z-t planes, effectively addressing anisotropies and non-uniformities in microbubble motion. This 2x2D approach improves the reconstruction of microbubbles’ dynamics by interpolating along both x and z directions. The method was tested on rat brain and ratkidney datasets recorded at 1kHz, demonstrating relaxing the frame rate to 100 Hz (using the first strategy) and a reduction in acquisition time by a factor of 3 to 4 (using the second strategy) while maintaining SR image quality comparable to the original uncompressed data, including density and velocity maps.

Automated cardiac segmentation from two-dimensional (2D) echocardiographic images is a crucial step toward improving clinical diagnosis. Anatomical heterogeneity and inherent noise, however, present technical challenges and lower segmentation accuracy. The objective of this study is to propose a method for the automatic segmentation of the ventricular endocardium, the myocardium, and the left atrium, in order to accurately determine clinical indices. Specifically, we suggest using the recently introduced pixel-to-pixel Generative Adversarial Network (Pix2Pix GAN) model for accurate segmentation. To accomplish this, we integrate the backbone PatchGAN model for the discriminator and the UNET for the generator, for building the Pix2Pix GAN. The resulting model produces precisely segmented images, thanks to UNET’s capability for precise segmentation and PatchGAN’s capability for fine-grained discrimination. For the experimental validation, we use the Cardiac Acquisitions for Multi-structure Ultrasound Segmentation (CAMUS) dataset, which consists of echocardiographic images from 500 patients in 2-chamber (2CH) and 4-chamber (4CH) views at the end-diastolic (ED) and end-systolic (ES) phases. Similarly to state-of-the-art studies on the same dataset, we followed the same train-test splits. Our results demonstrate that the proposed GAN-based technique improves segmentation performance for clinical and geometrical parameters compared to the state-of-the-art methods. More precisely, throughout the ED and ES phases, the mean Dice values for the left ventricular endocardium reached 0.961 and 0.930 for 2CH, and 0.959 and 0.950 for 4CH, respectively. Furthermore, the average ejection fraction correlation and Mean Absolute Error obtained were 0.95 and 3.2ml for 2CH, and 0.98 and 2.1ml for 4CH, outperforming the state-of-the-art results.

Ultrasound localization microscopy (ULM) allows for the generation of super-resolved (SR) images of the vasculature by precisely localizing intravenously injected microbubbles. Although SR images may be useful for diagnosing and treating patients, their use in the clinical context is limited by the need for prolonged acquisition times and high frame rates. The primary goal of our study is to relax the requirement of high frame rates to obtain SR images. To this end, we propose a new time-efficient ULM (TEULM) pipeline built on a cutting-edge interpolation method. More specifically, we suggest employing Radial Basis Functions (RBFs) as interpolators to estimate the missing values in the 2-dimensional (2D) spatio-temporal structures. To evaluate this strategy, we first mimic the data acquisition at a reduced frame rate by applying a down-sampling (DS = 2, 4, 8, and 10) factor to high frame rate ULM data. Then, we up-sample the data to the original frame rate using the suggested interpolation to reconstruct the missing frames. Finally, using both the original high frame rate data and the interpolated one, we reconstruct SR images using the ULM framework steps. We evaluate the proposed TEULM using four in vivo datasets, a Rat brain (dataset A), a Rat kidney (dataset B), a Rat tumor (dataset C) and a Rat brain bolus (dataset D), interpolating at the in-phase and quadrature (IQ) level. Results demonstrate the effectiveness of TEULM in recovering vascular structures, even at a DS rate of 10 (corresponding to a frame rate of sub-100Hz). In conclusion, the proposed technique is successful in reconstructing accurate SR images while requiring frame rates of one order of magnitude lower than standard ULM.

To solve the problem of reduced image quality in plane wave imaging (PWI), coherent plane wave compounding (CPWC) has been introduced, based on a combination of plane wave images from several directions (i.e., with different angles). However, the number of angles needed to reach a reasonable image quality affects the maximum achievable frame rate in CPWC. In this study, we suggest reducing the tradeoff between the image quality and the frame rate in CPWC by employing two-dimensional (2D) interpolation based on radial basis functions. More specifically, we propose constructing a three-dimensional spatio-angular structure to integrate both spatial and angular information into the reconstruction prior to 2D interpolation. The rationale behind our proposal is to reduce the number of transmissions and then apply the 2D interpolation along the angle dimension to reconstruct the missing information corresponding to the angles not selected for CPWC imaging. To evaluate the proposed technique, we applied it to the PWI challenges in the medical ultrasound database. Results show that we can achieve 3× to 4× improvement in frame rate while maintaining acceptable image quality compared to the case of using all the angles.

To address the low image quality issue in plane wave imaging (PWI), coherent plane-wave compounding (CPWC) was developed based on PWI along multiple directions followed by a coherent combination of the beamformed images. The main challenge with this approach is that there is a trade-off between image quality and frame rate which depends on how many angles are used in CPWC. Our primary goal here is to improve this trade-off by decreasing the number of angles required for image reconstruction. Here, we propose a sparse CPWC using a robust transmission coding pattern (TCP) followed by applying 2D spatio-angular interpolation. This idea allows us to improve the frame rate while preserving the image quality. We apply our proposed technique to the plane wave imaging challenge in the medical ultrasound (PICMUS) database. Results show that we can achieve a 3-fold frame rate improvement while preserving image quality.

Brain–computer interfaces (BCI) have begun to revolutionize many aspects of human life, ranging from health to smart living and communication devices. Therefore, BCI technologies require accurate recognition and classification systems of the brain responses to a variety of motor imagery (MI) movements through electroencephalogram (EEG) signals. Nowadays, deep learning models have received a lot of interest for learning features and classifying numerous kinds of data. However, it has not been thoroughly investigated for EEG signal classification. In this study, to reach an improved performance in MI‐EEG signal classification, we proposed a new strategy based on continuous wavelet transform (CWT)‐based time‐frequency maps to generate two‐dimensional (2D) EEG images of adaptively reconstructed signals from the extracted multivariate mode decomposition (MVMD) modes. In the next step, the extracted signals are projected in the new space using the proposed multiclass common spatial pattern (MCCSP) filtering. The resulting images are then fed to the convolutional neural network (CNN) architectures (AlexNet and LeNet). The proposed framework has the benefit of achieving high classification accuracy to be attained even with a significant amount of input data. LeNet and AlexNet reach the best average accuracy rates of 95.33% and 93.66% on dataset 1 from BCI competition IV and the results on dataset 2a from competition IV are more promising than the current state of the arts. Our results depict that MI‐EEG task recognition using image classification approaches, along with CNNs, is as comparable or even superior to other existing traditional approaches and provides high potential for upcoming research.