A Comprehensive Evaluation of Generalizability of Deep Learning-Based Hi-C Resolution Improvement Methods

Hi-C is a widely used technique to study the 3D organization of the genome. Due to its high sequencing cost, most of the generated datasets are of a coarse resolution, which makes it impractical to study finer chromatin features such as Topologically Associating Domains (TADs) and chromatin loops. Multiple deep learning-based methods have recently been proposed to increase the resolution of these datasets by imputing Hi-C reads (typically called upscaling). However, the existing works evaluate these methods on either synthetically downsampled datasets, or a small subset of experimentally generated sparse Hi-C datasets, making it hard to establish their generalizability in the real-world use case. We present our framework-Hi-CY-that compares existing Hi-C resolution upscaling methods on seven experimentally generated low-resolution Hi-C datasets belonging to various levels of read sparsities originating from three cell lines on a comprehensive set of evaluation metrics. Hi-CY also includes four downstream analysis tasks, such as TAD and chromatin loops recall, to provide a thorough report on the generalizability of these methods. We observe that existing deep learning methods fail to generalize to experimentally generated sparse Hi-C datasets, showing a performance reduction of up to 57%. As a potential solution, we find that retraining deep learning-based methods with experimentally generated Hi-C datasets improves performance by up to 31%. More importantly, Hi-CY shows that even with retraining, the existing deep learning-based methods struggle to recover biological features such as chromatin loops and TADs when provided with sparse Hi-C datasets. Our study, through the Hi-CY framework, highlights the need for rigorous evaluation in the future. We identify specific avenues for improvements in the current deep learning-based Hi-C upscaling methods, including but not limited to using experimentally generated datasets for training.

Ink Material Selection and Optical Design Considerations in DLP 3D Printing

Digital light processing (DLP) 3D printing has become a powerful manufacturing tool for the fast fabrication of complex functional structures. The rapid progress in DLP printing has been linked to research on optical design factors and ink selection. This critical review highlights the main challenges in the DLP printing of photopolymerizable inks. The kinetics equations of photopolymerization reaction in a DLP printer are solved, and the dependence of curing depth on the process optical parameters and ink chemical properties are explained. Developments in DLP platform design and ink selection are summarized, and the roles of monomer structure and molecular weight on DLP printing resolution are shown by experimental data. A detailed guideline is presented to help engineers and scientists to select inks and optical parameters for fabricating functional structures for multi-material and 4D printing applications.

Deep-E Enhanced Photoacoustic Tomography Using Three-Dimensional Reconstruction for High-Quality Vascular Imaging

Linear-array-based photoacoustic computed tomography (PACT) has been widely used in vascular imaging due to its low cost and high compatibility with current ultrasound systems. However, linear-array transducers have inherent limitations for three-dimensional imaging due to the poor elevation resolution. In this study, we introduced a deep learning-assisted data process algorithm to enhance the image quality in linear-array-based PACT. Compared to our earlier study where training was performed on 2D reconstructed data, here, we utilized 2D and 3D reconstructed data to train the two networks separately. We then fused the image data from both 2D and 3D training to get features from both algorithms. The numerical and in vivo validations indicate that our approach can improve elevation resolution, recover the true size of the object, and enhance deep vessels. Our deep learning-assisted approach can be applied to translational imaging applications that require detailed visualization of vascular features.

Do Radiographic Assessments of Periodontal Bone Loss Improve with Deep Learning Methods for Enhanced Image Resolution?

Resolution plays an essential role in oral imaging for periodontal disease assessment. Nevertheless, due to limitations in acquisition tools, a considerable number of oral examinations have low resolution, making the evaluation of this kind of lesion difficult. Recently, the use of deep-learning methods for image resolution improvement has seen an increase in the literature. In this work, we performed two studies to evaluate the effects of using different resolution improvement methods (nearest, bilinear, bicubic, Lanczos, SRCNN, and SRGAN). In the first one, specialized dentists visually analyzed the quality of images treated with these techniques. In the second study, we used those methods as different pre-processing steps for inputs of convolutional neural network (CNN) classifiers (Inception and ResNet) and evaluated whether this process leads to better results. The deep-learning methods lead to a substantial improvement in the visual quality of images but do not necessarily promote better classifier performance.

Second harmonic generation microscopy using pixel reassignment

Second harmonic generation (SHG) microscopy is expected to be a powerful tool for observing the cellular-level functionality and morphology information of thick tissue owe to its unique imaging properties. However, the maximum attainable resolution obtainable by SHG microscopy is limited by the use of long-wavelength, near-infrared excitation. In this paper, we report the use of pixel reassignment to improve the spatial resolution of SHG microscopy. The SHG signal is imaged onto a position-sensitive camera, instead of a point detector typically used in conventional SHG microscope. The data processing is performed through pixel reassignment and subsequent deblurring operation. We present the basic principle and a rigorous theoretical model for SHG microscopy using pixel reassignment (SHG-PR). And for the first time, the optimal reassignment factor for SHG-PR is derived based on the coherent characteristics and the dependence of wavelength in SHG microscopy. To evaluate the spatial resolution improvement, images of nano-beads separated by different distances and of a microtubule array have been simulated. We gain about a 1.5-fold spatial resolution enhancement compared to conventional SHG microscopy. When a further deblurring operation is implemented, this method allows for a total spatial resolution enhancement of about 1.87. Additionally, we demonstrate the validity of SHG-PR for raw data with noise. LAY DESCRIPTION: Second harmonic generation (SHG) microscopy has emerged as a powerful imaging technique in clinical diagnostics and biological research. SHG microscopy is label-free and provides intrinsic optical sectioning for three-dimensional (3D) imaging. However, a near-infrared excitation wavelength results a restriction in the maximum attainable spatial resolution of SHG microscopy. In this paper, we present a simple resolution-enhanced SHG imaging method, SHG microscopy using pixel reassignment (SHG-PR). We demonstrate a rigorous theoretical model for SHG-PR and derive the optimal reassignment factor. The simulation result shows the clear improvement of the image resolution and contrast in the SHG-PR after deblurring operation. The FWHM value of single microtubule shows that SHG-PR enables a spatial resolution enhancement by a factor of 1.5, compared to conventional SHG microscopy. After a proper deblurring operation, this method allows for a total spatial resolution enhancement of about 1.87. The improvements of spatial resolution and contrast are still valid for raw data with noise. It is expected that this method can contribute towards new insights in unstained tissue morphology, interaction of cells, and diseases diagnosis.

Sensitivity and Resolution Improvement in RGBW Color Filter Array Sensor

Recently, several red-green-blue-white (RGBW) color filter arrays (CFAs), which include highly sensitive W pixels, have been proposed. However, RGBW CFA patterns suffer from spatial resolution degradation owing to the sensor composition having more color components than the Bayer CFA pattern. RGBW CFA demosaicing methods reconstruct resolution using the correlation between white (W) pixels and pixels of other colors, which does not improve the red-green-blue (RGB) channel sensitivity to the W channel level. In this paper, we thus propose a demosaiced image post-processing method to improve the RGBW CFA sensitivity and resolution. The proposed method decomposes texture components containing image noise and resolution information. The RGB channel sensitivity and resolution are improved through updating the W channel texture component with those of RGB channels. For this process, a cross multilateral filter (CMF) is proposed. It decomposes the smoothness component from the texture component using color difference information and distinguishes color components through that information. Moreover, it decomposes texture components, luminance noise, color noise, and color aliasing artifacts from the demosaiced images. Finally, by updating the texture of the RGB channels with the W channel texture components, the proposed algorithm improves the sensitivity and resolution. Results show that the proposed method is effective, while maintaining W pixel resolution characteristics and improving sensitivity from the signal-to-noise ratio value by approximately 4.5 dB.