Image denoising is a crucial preprocessing step in various computer vision applications, yet remains challenging due to the diverse nature of noise and the trade-off between denoising efficacy and computational cost. This paper introduces MPDenoiseNet, a multi-path, resource-efficient deep learning network for image denoising, designed to handle a range of noise types including Bernoulli, Poisson, Salt-and-Pepper, and Gaussian noise, while being more computationally efficient than the transformer based, noise type agnostic restoration models. MPDenoiseNet architecture integrates parallel paths of Selective Convolutional (SeConv) blocks, adept at denoising Salt-and-Pepper corrupted images, and novel Anisotropic Diffusion blocks, designed to smooth images while preserving structural details. The features extracted from these paths are fused and further refined by Transformer blocks, inspired by Restormer, to capture long-range pixel interactions. Finally, an asymmetric U-Net AutoEncoder with skip connections module serves as a post-processing stage to enhance perceptual quality. MPDenoiseNet manages to reduce dependence on computationally expensive transformer blocks, using significantly fewer of them than the current state-of-the-art models and replacing them with more specialized lightweight blocks. We demonstrate MPDenoiseNet denoising performance is competitive with state-of-the-art methods like Restormer, while exhibiting efficient resource utilization in terms of both runtime and memory consumption.

Cervical cancer (CC) is considered a significant global health threat to women, therefore there is a need for personalized treatment strategies based on individual-specific gene expression patterns to enhance recovery and survival rates. Although a few studies have linked bisphosphoglycerate mutase (BPGM) expression with CC, its precise role in CC progression remains unclear. In this study, we conducted an integrated analysis for both bulk and single-cell RNA sequencing data to investigate the involvement of BPGM in CC. On the bulk RNA level, the Wilcoxon test result showed a significant upregulation of BPGM levels in CC tissue samples compared to the control samples, and particularly BPGM was found in the cancer cell subgroup within tumors on the single-cell RNA (scRNA) level. Furthermore, the survival analysis for BPGM level and its highly correlated genes indicated that, BPGM, egl-9 family hypoxia inducible factor 3 (EGLN3) as well as the Sad1 and UNC84 domain containing 1 (SUN1) could be predicting markers for the survival probabilities of CC patients with a significant p-value lower than 0.05. The pathway enrichment analysis for the BPGM and its correlated genes revealed various CC-related pathways such as PI3K-Akt and AMPK signalling pathways, which are very relevant for disease progression. Overall, these findings suggest that BPGM and its correlated genes could serve as a potential prognostic biomarker for CC.

More than a decade has passed since optics and genetics came together and lead to the emerging technologies of optogenetics. The advent of light-sensitive opsins made it possible to optically trigger the neurons into activation or inhibition by using visible light. In this work we integrated virtual reality optical system with a high radiance high density microdisplay to implement a wearable device for Retinitis Pigmentosa (RP) patients. We built system to emulate the human eye and light attenuation while it passes through the eye to the retina. The average light intensity on the surface of the retina after the optical system is 1.35mW/mm2which is above the ChR-2 stimulation level (0.7mW/mm2). The number of pixels which give light intensity above the stimulation threshold is 7128 pixel out of 8100 pixel in the microdisplay. The average efficiency of the optical system is 5.75%. So, the proposed wearable device will be able to restore vision for RP patients without the need for invasive implantable device by using optogenetics stimulation.

Cardiovascular disease is the main cause of death worldwide. Magnetic resonance imaging (MRI) has been considered as a noninvasive technique for characterizing myocardial tissues. Specifically, T2 mapping technique has been demonstrated to be an excellent tool to detect myocardial edema by estimation of the transverse relaxation time constant (T2) of myocardial tissue. However, there are several factors that could influence the analysis technique and affect the estimating T2 value. These factors include the type of exponential fitting model, which either be single exponential or exponential plus constant fitting model, the signal intensity estimation method, which either be Average, Median or Mapping method and signal-to-noise ratio (SNR) level of the T2-weighted images. In this paper, we discuss the effect of these factors on T2 estimation using a numerical phantom, T2 calibrated phantoms with different T2 values, and human subjects. The results of numerical phantom showed that the average percentage error for T2 measurement when using the single exponential fitting model reached 0.6 %, 0.4 %, and 5.9 % for the Average, Median, and Mapping methods, respectively. However, the exponential plus constant fitting model resulted in high average percentage error among the three signal intensity estimation methods (above 26 %). The experiments of calibrated phantom resulted in high correlation (R2> 0.99) between the estimated and reference T2 values when using the single exponential fitting model compared to low correlation (R2<; 0.73) when using the exponential plus constant fitting model at high SNR levels. Finally, the results of human subjects were in agreement with both numerical and calibrated phantoms. Based on the results of this paper, the single exponential fitting model with the median estimation method is the preferable analysis technique for T2 measurement as it results in a lower error for T2 measurements at low SNR levels and higher correlation values at high SNR levels compared to the other analysis techniques.

Portable image processing systems require algorithms which reduce the power consumption while maintaining video rate operation. This paper describes how splitting the data stream into multiple processing pipelines can reduce power consumption in contrast to the traditional spatial (pipeline) parallel processing technique. Real-time image processing system functions (Sobel filters and anisotropic diffusion) were implemented to show the principle of the technique. Our results show that partitioning the image into multiple sections gives increasing benefits with shorter processing times, and up to 45% drop in the power consumption.