Federated Learning (FL) is an advanced technology that tackles the challenge of blocked data arising from privacy concerns, enabling the training of deep learning models without the need for data sharing. However, FL faces difficulties with heterogeneous data and limited annotations in medical image segmentation. Motivated by this discovery, this study proposes a novel hybrid-supervised federated learning method (FedSLAG) that explores various types of annotations in medical imaging. To focus more on weakly-supervised and unsupervised scenarios within hybrid-supervised learning, a federated Gaussian enhancement module was proposed for heterogeneous sparse annotations (points and scribbles). The feature extraction module combines the features of multiple weakly-supervised clients and establishes the correlation of similar pixels, thus making up for the deficiency of scarce annotations and the insufficient feature extraction capability of a single machine. Then, a two-stage broadcast mechanism based on supervision sparsity was proposed to alleviate optimization deviation in local models. Experiments on breast tumor and skin lesion segmentation tasks demonstrate significant efficiency gains as well as highly competitive segmentation accuracy in many hybrid-supervised situations. Our codes are available at: https://github.com/TrivenDev/FedSLAG.

Accurate tumor segmentation in medical images plays an important role in clinical diagnosis and disease analysis. However, medical images usually have great complexity, such as low contrast of computed tomography (CT) or low spatial resolution of positron emission tomography (PET). In the actual radiotherapy plan, multimodal imaging technology, such as PET/CT, is often used. PET images provide basic metabolic information and CT images provide anatomical details. In this paper, we propose a 3D PET/CT tumor co-segmentation framework based on active contour model. First, a new edge stop function (ESF) based on PET image and CT image is defined, which combines the grayscale standard deviation information of the image and is more effective for blurry medical image edges. Second, we propose a background subtraction model to solve the problem of uneven grayscale level in medical images. Apart from that, the calculation format adopts the level set algorithm based on the additive operator splitting (AOS) format. The solution is unconditionally stable and eliminates the dependence on time step size. Experimental results on a dataset of 50 pairs of PET/CT images of non-small cell lung cancer patients show that the proposed method has a good performance for tumor segmentation.

This paper proposes a hybrid active contour model for tumor co-segmentation from PET/CT images. We incorporate the foreground of the CT image and the background of the PET image into a co-segmentation framework to establish a new energy functional. Different from existing methods, the proposed model incorporates an edge stopping function based on PET images. The proposed method has been evaluated on a data set of 50 pairs of PET/CT images of non-small cell lung cancer patients and compared with other single-mode segmentation methods and co-segmentation methods. Experimental results show that our model is more robust than other strategies in complex backgrounds.

Head and neck (H&N) cancers are among the most common cancers worldwide (5th leading cancer by incidence). Accurate segmentation of H&N tumors can improve the early diagnosis rate of cancers for timely treatment. H&N tumor segmentation challenge is the equidensity between the tumor and surrounding tissues, which shows low contrast in CT. In contrast, PET images can reflect the distinction between the lesion region and normal tissue through metabolic activity but show low spatial resolution. With the underlying assumption that each modality contains complementary information, we introduce a novel L2-Norm Scaled Transformer (NSTR) multi-modal segmentation method in PET-CT images. The proposed network comprises the Embedding block, L2-Norm Transformer blocks, 3D Deformable down-sampling blocks, and Feature fusion module, which can fully exploit the high sensitivity of PET images to tumors and the anatomical information of CT images. Our method proposes a powerful 3D fusion network that uses a U-shaped structure to exploit complementary features of different models at multiple scales to increase the cubical representations between different modalities. We conducted a comprehensive experimental analysis on the HECKTOR PET-CT dataset. The results indicated NSTR has powerful featured representation capability and surpasses the state-of-the-art H&N tumor segmentation methods in DSC, Jaccard, RVD, and HD95. (our code will be publicly available soon).

For data analysis, learning treatment rules in stratified medicine require the optimization of multiple responses. A common approach is to use a multi-objective function to find the optimal setting of the controllable factors. For patients, the optimal setting is a treatment regimen that yields the optimal value of potential responses. However, subclasses of patients are often stratified by their covariates. Thus, this paper proposes a new model called constrained optimization for stratified treatment rules (COSTAR) with multiple responses. This model incorporates covariates to build separate models for optimal responses and stratifies the patients with the balancing score from covariates. The optimal solution enables us to choose the optimal treatment for each subclass of patients. Theoretical results guarantee the identifiability of the solutions with conditional optimal values of multiple responses from survival probabilities. Examples of experiments with factorial designs and survival data validate the efficacy of the proposed method. The results suggest that this method improves the significance of the parameters and the adjusted R2 in fitting on the primary response, while the unsupervised clustering method (i.e., k-means) does not. This method, with the fitting model, is more interpretable than the conventional method and provides optimal treatment rules for stratified patients.

Prior learning is a fundamental problem in the field of image processing. In this paper, we conduct a detailed study on (1) how to model and learn the prior of the image patch group, which consists of a group of non-local similar image patches, and (2) how to apply the learned prior to the whole image denoising task. To tackle the first problem, we propose a new prior model named Group Sparsity Mixture Model (GSMM). With the bilateral matrix multiplication, the GSMM can model both the local feature of a single patch and the relation among non-local similar patches, and thus it is very suitable for patch group based prior learning. This is supported by the parameter analysis which demonstrates that the learned GSMM successfully captures the inherent strong sparsity embodied in the image patch group. Besides, as a mixture model, GSMM can be used for patch group classification. This makes the image denoising method based on GSMM capable of processing patch groups flexibly. To tackle the second problem, we propose an efficient and effective patch group based image denoising framework, which is plug-and-play and compatible with any patch group prior model. Using this framework, we construct two versions of GSMM based image denoising methods, both of which outperform the competing methods based on other prior models, e.g., Field of Experts (FoE) and Gaussian Mixture Model (GMM). Also, the better version is competitive with the state-of-the-art model based method WNNM with about ×8 faster average running speed.

Images with weak contrast, overlapped noise and texture of the object and background make many PDE based methods disabled. To address these problems, this paper presents a novel combined multi-scale variational framework level set segmentation model. Its level set formulation consists edge-based term, region-based term and shape constraint term. The edge-based term is constructed using a newly defined edge stopping function. The region-based term is derived from parameter-free Gaussian probability density function (pdf) and multiple Gaussian kernel are used to gray equalization. The shape constraint term is used to constrain contour evolution at different scales of image pyramid. For an intrinsic smoothing segmentation contours, the level set function is explicitly represented by B-spline basis functions. Finally, a convolution is used during the energy minimization. Experimental results on synthetic and real images validate the robustness and high accuracy boundaries detection for low contrast, noise and texture images.