Ali Mahloojifar

Segmentation is one of the most significant steps in image processing. Segmenting an image is a technique that makes it possible to separate a digital image into various areas based on the different characteristics of pixels in the image. In particular, the segmentation of breast ultrasound images is widely used for cancer identification. As a result of image segmentation, it is possible to make early diagnoses of a diseases via medical images in a very effective way. Due to various ultrasound artifacts and noises, including speckle noise, low signal-to-noise ratio, and intensity heterogeneity, the process of accurately segmenting medical images, such as ultrasound images, is still a challenging task. In this paper, we present a new method to improve the accuracy and effectiveness of breast ultrasound image segmentation. More precisely, we propose a neural network (NN) based on U-Net and an encoder-decoder architecture. By taking U-Net as the basis, both the encoder and decoder parts are developed by combining U-Net with other deep neural networks (Res-Net and MultiResUNet) and introducing a new approach and block (Co-Block), which preserve as much as possible the low-level and the high-level features. The designed network is evaluated using the Breast Ultrasound Images (BUSI) Dataset. It consists of 780 images, and the images are categorized into three classes, which are normal, benign, and malignant. According to our extensive evaluations on a public breast ultrasound dataset, the designed network segments the breast lesions more accurately than other state-of-the-art deep learning methods. With only 8.88 M parameters, our network (CResU-Net) obtained 82.88%, 77.5%, 90.3%, and 98.4% in terms of Dice similarity coefficients (DSC), intersection over union (IoU), area under curve (AUC), and global accuracy (ACC), respectively, on the BUSI dataset.

The Computer_Aided Diagnosis (CAD) systems facilitate accurate diagnosis of diseases. The development of CADs by leveraging third generation neural network, namely, Spiking Neural Network (SNN), is essential to utilize of the benefits of SNNs, such as their event_driven processing, parallelism, low power consumption, and the ability to process sparse temporal_spatial information. However, Deep SNN as a deep learning model faces challenges with unreliability. To deal with unreliability challenges due to inability to quantify the uncertainty of the predictions, we proposed a deep Bayesian Convolutional Spiking Neural Network based_CADs with uncertainty_aware module. In this study, the Monte Carlo Dropout method as Bayesian approximation is used as an uncertainty quantification method. This method was applied to several medical image classification tasks. Our experimental results demonstrate that our proposed model is accurate and reliable and will be a proper alternative to conventional deep learning for medical image classification.

Breast cancer is a heterogeneous disease with different molecular subtypes, clinical behavior, treatment responses as well as survival outcomes. The development of a reliable, accurate, available and inexpensive method to predict the molecular subtypes using medical images plays an important role in the diagnosis and prognosis of breast cancer. Recently, deep learning methods have shown good performance in the breast cancer classification tasks using various medical images. Despite all that success, classical deep learning cannot deliver the predictive uncertainty. The uncertainty represents the validity of the predictions. Therefore, the high predicted uncertainty might cause a negative effect in the accurate diagnosis of breast cancer molecular subtypes. To overcome this, uncertainty quantification methods are used to determine the predictive uncertainty. Accordingly, in this study, we proposed an uncertainty-aware Bayesian deep learning model using the full mammogram images. In addition, to increase the performance of the multi-class molecular subtype classification task, we proposed a novel hierarchical classification strategy, named the two-stage classification strategy. The separate AUC of the proposed model for each subtype was 0.71, 0.75 and 0.86 for HER2-enriched, luminal and triple-negative classes, respectively. The proposed model not only has a comparable performance to other studies in the field of breast cancer molecular subtypes prediction, even using full mammography images, but it is also more reliable, due to quantify the predictive uncertainty.

Purpose Phased subarray imaging (PSA) was previously proposed to extend the receive aperture length. Using overlapped subarrays as transmitters in PSA leads to decrement of sidelobe levels of the overall beam compared to full phased array imaging (PHA). This paper proposes an adaptive compounding of subarray images in PSA to improve both the resolution and contrast compared with PHA. Method Adaptive apodization (ADAP) is defined proportional to the beamformed responses of subarrays such that the overall energy after compounding is minimized. Results The simulation and experimental results validate the performance of applying ADAP in PSA. The full width at half maximum (FWHM) at a depth of 30 mm in the proposed PSA is about 0.2 mm, compared to a FWHM of 0.6 mm with PHA imaging. Measuring the contrast ratio index shows that the ADAP method also improves the contrast in PSA imaging at least 25% compared to PHA imaging. Conclusion Applying the proposed ADAP, besides conventional compounding in PSA imaging, leads to improvement of both the resolution and contrast compared to PHA imaging.

In Photoacoustic imaging, Delay-and-Sum (DAS) algorithm is the most commonly used beamformer. However, it leads to a low resolution and high level of sidelobes. Delay-Multiply-and-Sum (DMAS) was introduced to provide lower sidelobes compared to DAS. In this paper, to improve the resolution and sidelobes of DMAS, a novel beamformer is introduced using Eigenspace-Based Minimum Variance (EIBMV) method combined with DMAS, namely EIBMV-DMAS. It is shown that expanding the DMAS algebra leads to several terms which can be interpreted as DAS. Using the EIBMV adaptive beamforming instead of the existing DAS (inside the DMAS algebra expansion) is proposed to improve the image quality. EIBMV-DMAS is evaluated numerically and experimentally. It is shown that EIBMV-DMAS outperforms DAS, DMAS and EIBMV in terms of resolution and sidelobes. In particular, at the depth of 11 mm of the experimental images, EIBMV-DMAS results in about 113 dB and 50 dB sidelobe reduction, compared to DMAS and EIBMV, respectively. At the depth of 7 mm, for the experimental images, the quantitative results indicate that EIBMV-DMAS leads to improvement in Signal-to-Noise Ratio (SNR) of about 75% and 34%, compared to DMAS and EIBMV, respectively.

Photoacoustic imaging (PAI) is a novel medical imaging modality that uses the advantages of the spatial resolution of ultrasound imaging and the high contrast of pure optical imaging. Analytical algorithms are usually employed to reconstruct the photoacoustic (PA) images as a result of their simple implementation. However, they provide a low accurate image. Model-based (MB) algorithms are used to improve the image quality and accuracy while a large number of transducers and data acquisition are needed. In this paper, we have combined the theory of compressed sensing (CS) with MB algorithms to reduce the number of transducer. Smoothed version of L0-norm (SL0) was proposed as the reconstruction method, and it was compared with simple iterative reconstruction (IR) and basis pursuit. The results show that S$\ell_0$ provides a higher image quality in comparison with other methods while a low number of transducers were. Quantitative comparison demonstrates that, at the same condition, the SL0 leads to a peak-signal-to-noise ratio for about two times of the basis pursuit.

In Photoacoustic imaging (PA), Delay-and-Sum (DAS) beamformer is a common beamforming algorithm having a simple implementation. However, it results in a poor resolution and high sidelobes. To address these challenges, a new algorithm namely Delay-Multiply-and-Sum (DMAS) was introduced having lower sidelobes compared to DAS. To improve the resolution of DMAS, a novel beamformer is introduced using Minimum Variance (MV) adaptive beamforming combined with DMAS, so-called Minimum Variance-Based DMAS (MVB-DMAS). It is shown that expanding the DMAS equation results in multiple terms representing a DAS algebra. It is proposed to use the MV adaptive beamformer instead of the existing DAS. MVB-DMAS is evaluated numerically and experimentally. In particular, at the depth of 45 mm MVB-DMAS results in about 31 dB, 18 dB and 8 dB sidelobes reduction compared to DAS, MV and DMAS, respectively. The quantitative results of the simulations show that MVB-DMAS leads to improvement in full-width-half-maximum about 96 %, 94 % and 45 % and signal-to-noise ratio about 89 %, 15 % and 35 % compared to DAS, DMAS, MV, respectively. In particular, at the depth of 33 mm of the experimental images, MVB-DMAS results in about 20 dB sidelobes reduction in comparison with other beamformers.

Delay-and-Sum (DAS) beamformer is the most common beamforming algorithm in Photoacoustic imaging (PAI) due to its simple implementation and real time imaging. However, it provides poor resolution and high levels of sidelobe. A new algorithm named Delay-Multiply-and-Sum (DMAS) was introduced. Using DMAS leads to lower levels of sidelobe compared to DAS, but resolution is not satisfying yet. In this paper, a novel beamformer is introduced based on the combination of Minimum Variance (MV) adaptive beamforming and DMAS, so-called Minimum Variance-Based DMAS (MVB-DMAS). It is shown that expanding the DMAS equation leads to some terms which contain a DAS equation. It is proposed to use MV adaptive beamformer instead of existing DAS inside the DMAS algebra expansion. MVB-DMAS is evaluated numerically compared to DAS, DMAS and MV and Signal-to-noise ratio (SNR) metric is presented. It is shown that MVB-DMAS leads to higher image quality and SNR for about 13 dB, 3 dB and 2 dB in comparison with DAS, DMAS and MV, respectively.