Gated Skip-Connection Network with Adaptive Upsampling for Retinal Vessel Segmentation

GLGAN-VC: A Guided Loss-Based Generative Adversarial Network for Many-to-Many Voice Conversion

Many-to-many voice conversion (VC) is a technique aimed at mapping speech features between multiple speakers during training and transferring the vocal characteristics of one source speaker to another target speaker, all while maintaining the content of the source speech unchanged. Existing research highlights a notable gap between the original and generated speech samples in terms of naturalness within many-to-many VC. Therefore, there is substantial room for improvement in achieving more natural-sounding speech samples for both parallel and nonparallel VC scenarios. In this study, we introduce a generative adversarial network (GAN) system with a guided loss (GLGAN-VC) designed to enhance many-to-many VC by focusing on architectural improvements and the integration of alternative loss functions. Our approach includes a pair-wise downsampling and upsampling (PDU) generator network for effective speech feature mapping (FM) in multidomain VC. In addition, we incorporate an FM loss to preserve content information and a residual connection (RC)-based discriminator network to improve learning. A guided loss (GL) function is introduced to efficiently capture differences in latent feature representations between source and target speakers, and an enhanced reconstruction loss is proposed for better contextual information preservation. We evaluate our model on various datasets, including VCC 2016, VCC 2018, VCC 2020, and an emotional speech dataset (ESD). Our results, based on both subjective and objective evaluation metrics, demonstrate that our model outperforms state-of-the-art (SOTA) many-to-many GAN-based VC models in terms of speech quality and speaker similarity in the generated speech samples.

A modified U-net with graph representation for dose prediction in esophageal cancer radiotherapy plans

The manual design of esophageal cancer radiotherapy plan is time-consuming and labor-intensive. Automatic planning (AP) is prevalent nowadays to increase physicists' work efficiency. Because of the intuitiveness of dose distribution in AP evaluation, obtaining reasonable dose prediction provides effective guarantees to generate a satisfactory AP. Existing fully convolutional network-based methods for predicting dose distribution in esophageal cancer radiotherapy plans often capture features in a limited receptive field. Additionally, the correlations between voxel pairs are often ignored. This work modifies the U-net architecture and exploits graph convolution to capture long-range information for dose prediction in esophageal cancer plans. Meanwhile, attention mechanism gets correlations between planning target volume (PTV) and organs at risk, and adaptively learns their feature weights. Finally, a novel loss function that considers features between voxel pairs is used to highlight the predictions. 152 subjects with prescription doses of 50 Gy or 60 Gy are collected in this study. The mean absolute error and standard deviation of conformity index, homogeneity index, and max dose for PTV achieved by the proposed method are 0.036 ± 0.030, 0.036 ± 0.027, and 0.930 ± 1.162, respectively, which outperform other state-of-the-art models. The superior performance demonstrates that our proposed method has great potential for AP generation.

Improved UNet with Attention for Medical Image Segmentation

Medical image segmentation is crucial for medical image processing and the development of computer-aided diagnostics. In recent years, deep Convolutional Neural Networks (CNNs) have been widely adopted for medical image segmentation and have achieved significant success. UNet, which is based on CNNs, is the mainstream method used for medical image segmentation. However, its performance suffers owing to its inability to capture long-range dependencies. Transformers were initially designed for Natural Language Processing (NLP), and sequence-to-sequence applications have demonstrated the ability to capture long-range dependencies. However, their abilities to acquire local information are limited. Hybrid architectures of CNNs and Transformer, such as TransUNet, have been proposed to benefit from Transformer's long-range dependencies and CNNs' low-level details. Nevertheless, automatic medical image segmentation remains a challenging task due to factors such as blurred boundaries, the low-contrast tissue environment, and in the context of ultrasound, issues like speckle noise and attenuation. In this paper, we propose a new model that combines the strengths of both CNNs and Transformer, with network architectural improvements designed to enrich the feature representation captured by the skip connections and the decoder. To this end, we devised a new attention module called Three-Level Attention (TLA). This module is composed of an Attention Gate (AG), channel attention, and spatial normalization mechanism. The AG preserves structural information, whereas channel attention helps to model the interdependencies between channels. Spatial normalization employs the spatial coefficient of the Transformer to improve spatial attention akin to TransNorm. To further improve the skip connection and reduce the semantic gap, skip connections between the encoder and decoder were redesigned in a manner similar to that of the UNet++ dense connection. Moreover, deep supervision using a side-output channel was introduced, analogous to BASNet, which was originally used for saliency predictions. Two datasets from different modalities, a CT scan dataset and an ultrasound dataset, were used to evaluate the proposed UNet architecture. The experimental results showed that our model consistently improved the prediction performance of the UNet across different datasets.

CPMF-Net: Multi-Feature Network Based on Collaborative Patches for Retinal Vessel Segmentation

As an important basis of clinical diagnosis, the morphology of retinal vessels is very useful for the early diagnosis of some eye diseases. In recent years, with the rapid development of deep learning technology, automatic segmentation methods based on it have made considerable progresses in the field of retinal blood vessel segmentation. However, due to the complexity of vessel structure and the poor quality of some images, retinal vessel segmentation, especially the segmentation of Capillaries, is still a challenging task. In this work, we propose a new retinal blood vessel segmentation method, called multi-feature segmentation, based on collaborative patches. First, we design a new collaborative patch training method which effectively compensates for the pixel information loss in the patch extraction through information transmission between collaborative patches. Additionally, the collaborative patch training strategy can simultaneously have the characteristics of low occupancy, easy structure and high accuracy. Then, we design a multi-feature network to gather a variety of information features. The hierarchical network structure, together with the integration of the adaptive coordinate attention module and the gated self-attention module, enables these rich information features to be used for segmentation. Finally, we evaluate the proposed method on two public datasets, namely DRIVE and STARE, and compare the results of our method with those of other nine advanced methods. The results show that our method outperforms other existing methods.

Predicting narrow ureters before ureteroscopic lithotripsy with a neural network: a retrospective bicenter study

In some patients, the passage of semi-rigid ureteroscopes up the ureter is impossible due to narrow ureteral lumen. We established a neural network to predict the inability of the ureter to accommodate the semi-rigid ureteroscope and the need for active or passive dilatation using non-contrast computed tomography (CT) images. Data were collected retrospectively from two centers of 1989 eligible patients who underwent ureteroscopic lithotripsy with ureteral stones. Patients were categorized into two groups: control and narrow ureter. The network was designed and trained for predicting a narrow ureter during initial ureteroscopic lithotripsy, which integrated multi-scale features of the ureter. The predictive efficacy of neural networks DenseNet3D, ResNet3D, ResNet3D MC, and TimeSformer was compared. Furthermore, a previous ureteroscopy or a history of double-J stent placement, ureteral wall thickness and Hounsfield unit (HU) density of the ureter under the stone were compared. Model performance was assessed based on the accuracy, area under the receiver operating characteristic curve (AUC ROC), etc. The DenseNet3D-based network achieved an AUC ROC score of 0.884 and an accuracy of 85.29%, followed by the ResNet3D-based network, the ResNet3D MC-based network, and the TimeSformer-based network. The DenseNet3D-based network significantly outperformed other candidate predictors. Furthermore, the networks were validated in an external test set. Decision curve analysis showed the clinical utility of the neural network. The neural network provides an individualized preoperative prediction of narrow ureter based on non-contrast CT images, which could be employed as part of a surgical decision-making support system.

LightEyes: A Lightweight Fundus Segmentation Network for Mobile Edge Computing

Fundus is the only structure that can be observed without trauma to the human body. By analyzing color fundus images, the diagnosis basis for various diseases can be obtained. Recently, fundus image segmentation has witnessed vast progress with the development of deep learning. However, the improvement of segmentation accuracy comes with the complexity of deep models. As a result, these models show low inference speeds and high memory usages when deploying to mobile edges. To promote the deployment of deep fundus segmentation models to mobile devices, we aim to design a lightweight fundus segmentation network. Our observation comes from the fact that high-resolution representations could boost the segmentation of tiny fundus structures, and the classification of small fundus structures depends more on local features. To this end, we propose a lightweight segmentation model called LightEyes. We first design a high-resolution backbone network to learn high-resolution representations, so that the spatial relationship between feature maps can be always retained. Meanwhile, considering high-resolution features means high memory usage; for each layer, we use at most 16 convolutional filters to reduce memory usage and decrease training difficulty. LightEyes has been verified on three kinds of fundus segmentation tasks, including the hard exudate, the microaneurysm, and the vessel, on five publicly available datasets. Experimental results show that LightEyes achieves highly competitive segmentation accuracy and segmentation speed compared with state-of-the-art fundus segmentation models, while running at 1.6 images/s Cambricon-1A speed and 51.3 images/s GPU speed with only 36k parameters.