DM-CNN: Dynamic Multi-scale Convolutional Neural Network with uncertainty quantification for medical image classification.

Convolutional neural network (CNN) has promoted the development of diagnosis technology of medical images. However, the performance of CNN is limited by insufficient feature information and inaccurate attention weight. Previous works have improved the accuracy and speed of CNN but ignored the uncertainty of the prediction, that is to say, uncertainty of CNN has not received enough attention. Therefore, it is still a great challenge for extracting effective features and uncertainty quantification of medical deep learning models In order to solve the above problems, this paper proposes a novel convolutional neural network model named DM-CNN, which mainly contains the four proposed sub-modules : dynamic multi-scale feature fusion module (DMFF), hierarchical dynamic uncertainty quantifies attention (HDUQ-Attention) and multi-scale fusion pooling method (MF Pooling) and multi-objective loss (MO loss). DMFF select different convolution kernels according to the feature maps at different levels, extract different-scale feature information, and make the feature information of each layer have stronger representation ability for information fusion HDUQ-Attention includes a tuning block that adjust the attention weight according to the different information of each layer, and a Monte-Carlo (MC) dropout structure for quantifying uncertainty MF Pooling is a pooling method designed for multi-scale models, which can speed up the calculation and prevent overfitting while retaining the main important information Because the number of parameters in the backbone part of DM-CNN is different from other modules, MO loss is proposed, which has a fast optimization speed and good classification effect DM-CNN conducts experiments on publicly available datasets in four areas of medicine (Dermatology, Histopathology, Respirology, Ophthalmology), achieving state-of-the-art classification performance on all datasets. DM-CNN can not only maintain excellent performance, but also solve the problem of quantification of uncertainty, which is a very important task for the medical field. The code is available: https://github.com/QIANXIN22/DM-CNN.

A Network with Composite Loss and Parameter-free Chunking Fusion Block for Super-Resolution MR Image.

MRI is often influenced by many factors, and single image super-resolution (SISR) based on a neural network is an effective and cost-effective alternative technique for the high-resolution restoration of low-resolution images. However, deep neural networks can easily lead to overfitting and make the test results worse. The network with a shallow training network is difficult to fit quickly and cannot completely learn training samples. To solve the above problems, a new end-to-end super-resolution (SR) method is proposed for magnetic resonance (MR) images. Firstly, in order to better fuse features, a parameter-free chunking fusion block (PCFB) is proposed, which can divide the feature map into n branches by splitting channels to obtain parameter-free attention. Secondly, the proposed training strategy including perceptual loss, gradient loss, and L1 loss has significantly improved the accuracy of model fitting and prediction. Finally, the proposed model and training strategy take the super-resolution IXISR dataset (PD, T1, and T2) as an example to compare with the existing excellent methods and obtain advanced performance. A large number of experiments have proved that the proposed method performs better than the advanced methods in highly reliable measurement.

EHDFL: Evolutionary hybrid domain feature learning based on windowed fast Fourier convolution pyramid for medical image classification.

With the development of modern medical technology, medical image classification has played an important role in medical diagnosis and clinical practice. Medical image classification algorithms based on deep learning emerge in endlessly, and have achieved amazing results. However, most of these methods ignore the feature representation based on frequency domain, and only focus on spatial features. To solve this problem, we propose a hybrid domain feature learning (HDFL) module based on windowed fast Fourier convolution pyramid, which combines the global features with a wide range of receptive fields in frequency domain and the local features with multiple scales in spatial domain. In order to prevent frequency leakage, we construct a Windowed Fast Fourier Convolution (WFFC) structure based on Fast Fourier Convolution (FFC). In order to learn hybrid domain features, we combine ResNet, FPN, and attention mechanism to construct a hybrid domain feature learning module. In addition, a super-parametric optimization algorithm is constructed based on genetic algorithm for our classification model, so as to realize the automation of our super-parametric optimization. We evaluated the newly published medical image classification dataset MedMNIST, and the experimental results show that our method can effectively learning the hybrid domain feature information of frequency domain and spatial domain.

HWA-SegNet: Multi-channel skin lesion image segmentation network with hierarchical analysis and weight adjustment.

Convolutional neural networks (CNNs) show excellent performance in accurate medical image segmentation. However, the characteristics of sample with small size and insufficient feature expression, irregular shape of the segmented target and inaccurate judgment of edge texture have always been problems to be faced in the field of skin lesion image segmentation. Therefore, in order to solve these problems, discrete Fourier transform (DFT) is introduced to enrich the input data and a CNN architecture (HWA-SegNet) is proposed in this paper. Firstly, DFT is improved to analyze the features of the skin lesions image, and multi-channel data is extended for each image. Secondly, a hierarchical dilated analysis module is constructed to understand the semantic features under multi-channel. Finally, the pre-prediction results are fine-tuned using a weight adjustment structure with fully connected layers to obtain higher accuracy prediction results. Then, 520 skin lesion images are tested on the ISIC 2018 dataset. Extensive experimental results show that our HWA-SegNet improve the average segmentation Dice Similarity Coefficient from 88.30% to 91.88%, Sensitivity from 89.29% to 92.99%, and Jaccard similarity index from 81.15% to 85.90% compared with U-Net. Compared with the State-of-the-Art method, the Jaccard similarity index and Specificity are close, but the Dice Similarity Coefficient is higher. The experimental data show that the data augmentation strategy based on improved DFT and HWA-SegNet are effective for skin lesion image segmentation.

Choroid Segmentation of Retinal OCT Images Based on CNN Classifier and l 2-l q Fitter.

Optical coherence tomography (OCT) is a noninvasive cross-sectional imaging technology used to examine the retinal structure and pathology of the eye. Evaluating the thickness of the choroid using OCT images is of great interests for clinicians and researchers to monitor the choroidal thickness in many ocular diseases for diagnosis and management. However, manual segmentation and thickness profiling of choroid are time-consuming which lead to low efficiency in analyzing a large quantity of OCT images for swift treatment of patients. In this paper, an automatic segmentation approach based on convolutional neural network (CNN) classifier and l 2-l q (0 < q < 1) fitter is presented to identify boundaries of the choroid and to generate thickness profile of the choroid from retinal OCT images. The method of detecting inner choroidal surface is motivated by its biological characteristics after light reflection, while the outer chorioscleral interface segmentation is transferred into a classification and fitting problem. The proposed method is tested in a data set of clinically obtained retinal OCT images with ground-truth marked by clinicians. Our numerical results demonstrate the effectiveness of the proposed approach to achieve stable and clinically accurate autosegmentation of the choroid.

Polarization distribution control of anisotropic electromagnetic Gaussian-Schell model beams on free propagation by exploiting correlation properties at the source plane.

When propagating in free space, the transversal distribution of the degree of polarization of an anisotropic electromagnetic Gaussian-Schell model (AEGSM) beam will generally undergo a complex evolution process. We find that this transversal distribution of the degree of polarization of an AEGSM beam can be controlled by exploiting the partial correlation properties of the source. The main research of our paper falls into two parts. First, the concept of analogical propagation of the transversal distribution of the degree of polarization is proposed, and the condition for an AEGSM beam having an analogical propagation is obtained. When an AEGSM beam is on analogical propagation, the distribution of the degree of polarization on any cross section of the beam is always similar to that on the source plane, except that the size of the distribution pattern will expand continuously as the propagation distance increases. Second, the far-field transversal distribution of the degree of polarization is considered, and the condition for the far-field transversal polarization distribution of an AEGSM beam to be always of circularly symmetric shape, no matter how complicated it is on the source, is obtained. Our research is expected to find applications in areas that make use of the polarization properties of random electromagnetic beams.

Aberration-induced intensity imbalance of alternating phase-shifting mask in lithographic imaging.

We discuss the aberration-induced intensity imbalance of alternating phase-shifting mask (Alt-PSM) in lithographic imaging, in contrast to numerous studies on mask-induced intensity imbalance. Based on the Hopkins theory of partial coherent imaging, a linear relationship between the intensity difference of adjacent peaks in an Alt-PSM image and even aberration is established by formulations and verified by numerical results. The application of the linear relationship is briefly discussed.

Even aberration measurement of lithographic projection optics based on intensity difference of adjacent peaks in alternating phase-shifting mask image.

We propose an in situ technique for measuring an even aberration of lithographic projection optics. By using the Hopkins theory of partially coherent imaging and the thick-mask model, the linear relationship between the intensity difference of adjacent peaks in an alternating phase-shifting mask image and an even aberration is established by equations and verified by numerical results. The sensitivity of measuring the even aberration of lithographic projection optics based on this linear relationship is analyzed, and the measurement mark is designed accordingly. Measurement performance of the present technique is evaluated using the lithographic simulator PROLITH, which shows that the present technique is capable of measuring the even aberration of lithographic projection optics with ultrahigh measurement accuracy.

Translational-symmetry alternating phase shifting mask grating mark used in a linear measurement model of lithographic projection lens aberrations.

A linear measurement model of lithographic projection lens aberrations is studied numerically based on the Hopkins theory of partially-coherent imaging and positive resist optical lithography (PROLITH) simulation. In this linearity model, the correlation between the mark's structure and its sensitivities to aberrations is analyzed. A method to design a mark with high sensitivity is proved and declared. By use of this method, a translational-symmetry alternating phase shifting mask (Alt-PSM) grating mark is redesigned with all of the even orders, +/-3rd and +/-5th order diffraction light missing. In the evaluation simulation, the measurement accuracies of aberrations prove to be enhanced apparently by use of the redesigned mark instead of the old ones.

Coma measurement by use of an alternating phase-shifting mask mark with a specific phase width.

The correlation between the coma sensitivity of the alternating phase-shifting mask (Alt-PSM) mark and the mark's structure is studied based on the Hopkins theory of partially coherent imaging and positive resist optical lithography (PROLITH) simulation. It is found that an optimized Alt-PSM mark with its phase width being two-thirds its pitch has a higher sensitivity to coma than Alt-PSM marks with the same pitch and the different phase widths. The pitch of the Alt-PSM mark is also optimized by PROLITH simulation, and the structure of p=1.92lambda/NA and pw=2p/3 proves to be with the highest sensitivity. The optimized Alt-PSM mark is used as a measurement mark to retrieve coma aberration from the projection optics in lithographic tools. In comparison with an ordinary Alt-PSM mark with its phase width being a half its pitch, the measurement accuracies of Z(7) and Z(14) apparently increase.

Coma measurement of projection optics in lithographic tools based on relative image displacements at multiple illumination settings.

In this paper, we propose a novel method for measuring the coma aberrations of lithographic projection optics based on relative image displacements at multiple illumination settings. The measurement accuracy of coma can be improved because the phase-shifting gratings are more sensitive to the aberrations than the binary gratings used in the TAMIS technique, and the impact of distortion on displacements of aerial image can be eliminated when the relative image displacements are measured. The PROLITH simulation results show that, the measurement accuracy of coma increases by more than 25% under conventional illumination, and the measurement accuracy of primary coma increases by more than 20% under annular illumination, compared with the TAMIS technique.