Class incremental medical image segmentation (CIMIS) aims to preserve knowledge of previously learned classes while learning new ones without relying on old-class annotations. However, existing methods 1) either adopt one-size-fits-all strategies that treat all spatial regions and feature channels equally, which may hinder the preservation of accurate old knowledge, 2) or focus solely on aligning local prototypes with global ones for old classes while overlooking their local representations in new data, leading to knowledge degradation. To mitigate the above issues, we propose Prototype-Guided Calibration Distillation (PGCD) and Dual-Aligned Prototype Distillation (DAPD) for CIMIS in this paper. Specifically, PGCD exploits prototype-to-feature similarity to calibrate class-specific distillation intensity in different spatial regions, effectively reinforcing reliable old knowledge and suppressing misleading cues from old classes. Complementarily, DAPD aligns the local prototypes of old classes extracted from the current model with both global historical prototypes and local prototypes, further enhancing segmentation performance on old categories. Comprehensive evaluations on two widely used multi-organ segmentation benchmarks demonstrate that our method outperforms current state-of-the-art methods, highlighting its robustness and generalization capabilities.

The Transformer architecture has achieved remarkable success in computer vision and natural language processing. However, its application to time series forecasting frequently results in performance degradation, occasionally underperforming even simple linear models. Prior analyses have predominantly attributed this limitation to suboptimal embedding designs that fail to construct a well-structured latent space capable of effectively capturing intricate temporal dependencies. Most existing architectures rely on linear embeddings for input mapping, but these transformations often fail to project raw time series data onto a high-dimensional manifold, resulting in latent representations that poorly capture complex temporal structures and lead to attention degeneration. To overcome these challenges, we propose Structured Latent Projection (SLP), an enhanced embedding method that generates a rich, structured latent space from raw time series data. By mapping input sequences to a high-dimensional manifold that captures multi-scale temporal dependencies and inter-variable interactions, SLP improves the efficiency and robustness of the self-attention mechanism. We integrate SLP into the Transformer architecture, resulting in a model called LatentBridge. Extensive experiments on 13 real-world datasets show that LatentBridge consistently achieves state-of-the-art performance in both long-term and short-term forecasting.

Inverse treatment planning is pivotal in tumor treatment planning. It enables the multi-objective optimization of radiation dose delivery, ensuring precise tumor targeting while sparing surrounding healthy tissues. This process often requires frequent parameter adjustments to achieve the desired balance between objectives, making it both labor-intensive and time-consuming. Deep reinforcement learning (DRL) provides an automated, model-based planning solution, aimed at reducing reliance on human expertise and enhancing the efficiency of objective parameter optimization. However, most current approaches apply DRL to inverse planning without fully leveraging the knowledge embedded in the continuous state-action space, defined by the coupling between nonstationary planning states and continuous decision variables. This may result in insufficient exploration and exploitation, leading to inefficient optimization. This work introduces an innovative action-guided DRL (AgDRL) approach for automatic inverse planning. Our goal is to enhance exploration and exploitation by leveraging insightful guidance from reward-guided actions. The implementation of AgDRL incorporates both exploitation and exploration in the action-state space. For exploitation, high-reward actions are employed as guidance to achieve the optimal action adjustment. For exploration, low-reward actions are recommended as training resets to explore a broader range of the latent state space. Quantitative and qualitative experiments are conducted in various settings to evaluate the proposed method. The results are assessed using DRL-related metrics (e.g. reward gains) and clinical-related measurements (e.g. dose-volume histograms, DVHs). Experimental results on a real-world rectal cancer dataset empirically demonstrate that the proposed AgDRL-based approach significantly improves optimization efficiency through a high-reward strategy while enhancing exploration diversity via a low-reward strategy, consistently outperforming the MatRad treatment planning optimization platform.

Magnetic resonance imaging (MRI)-guided radiotherapy (MRIgRT) integrates MRI with linear accelerators (MRI-linacs), enabling real-time motion management based on temporally resolved 2D MRI (cine-MRI). Current systems rely on template matching or deformable image registration for radiotherapy target (typically the gross tumor volume) localization, which allows beam gating. Further advances in localization could support more precise and efficient delivery methods. https://trackrad2025.grand-challenge.org/ was organized to provide a common dataset to benchmark algorithms for MRIgRT target tracking in 2D+t cine-MRI. Participants propagated target segmentation masks from an initialization frame across subsequent frames. The dataset comprised sagittal cine-MRI scans of 585 cancer patients undergoing radiotherapy at 0.35 T and 1.5 T MRI-linacs at six different institutions, with expert-annotated targets in 108 sequences. Target sites included the thorax (179 cases), abdomen (266 cases), and pelvis (140 cases). A total of 477 unlabeled and 50 labeled cases were provided for training purposes, 58 cases were kept private for preliminary testing (8) and final evaluation (50). The algorithms submitted by participants were executed on the challenge platform and assessed using metrics in three categories: geometric accuracy, surrogate dose accuracy and execution speed. Rankings were derived via a Rank-Then-Mean scheme. TrackRAD2025 attracted 148 registrations from 28 countries, 100 preliminary submissions and 24 final submissions from 14 teams. The top five methods achieved mean Dice similarity coefficients >0.87 and Euclidean center distances <2.1 mm, comparable to interobserver variability. Leading top five solutions featured foundation models with (4) or without (1) finetuning. Field strength had minimal effect on performance and tracking worked better for the pelvis with reduced motion amplitude compared to the thorax and abdomen cases, which achieved equivalent performance. TrackRAD2025 established a benchmark for MRIgRT tracking on multi-institutional cine-MRI data, highlighting foundation models as promising for clinical translation.

Propagation-based methods have drawn increasing research attention in interactive medical image segmentation. However, existing propagation-based methods face two significant challenges: 1) Due tothe continuous nature of anatomical structures within the organs and tumors throughout the volume, over-propagation is likely to occur as the propagation process reaches the end of structures, leadingto a degradation in segmentation performance. 2) During the multi-round refinement process, selecting the worst-segmented slice for refinement tends to hinder the optimization of segmentation results. To overcome these challenges, we propose the Discrepancy Aware Network (DANet), which includes a Discrepancy Learning Module (DLM) and employs a confidence loss to achieve accurate segmentation. Specifically, DLM captures the temporal-contextual discrepancy between previous and current slices, enabling the model to perceive the variations of the target. Furthermore, the confidence loss is responsible for regularizing the over-confident segmentation at the image level by estimating the target foreground. Additionally, we design a straightforward slice selection strategy to optimize the refinement process. Extensive experimental results on five public medical datasets demonstrate significant improvements over state-of-the-art methods (e.g., with +1.07% improvement on the MSD-Spleen dataset).

Fluence map optimization problem (FMOP) refers to the optimization of the intensity of radiation beams, which is a crucial part of intensity-modulated radiation therapy (IMRT). The FMOP is often considered as a multi-objective optimization problem due to the numerous treatment objectives that need to be met. This paper formulates FMOP as an unconstrained two-objective optimization problem that focuses on dose-volume constraints and specifically designed a multi-objective coevolutionary optimization algorithm named MOEA/FMOP. The MOEA/FMOP utilizes a cooperative strategy to process dual populations with two different resolutions of fluence maps. One high-resolution population focuses on convergence and fine-tuning, while the other roughly encoded one serves to improve global convergence and maintain diversity. The resolutions of the populations are encoded and initialized according to the clinical methodology. In comparison to conventional MOEAs, MOEA/FMOP outperforms over five real-world cancer cases (including prostate, rectum, liver, nasopharynx and breast cases) in terms of hyper volume (HV) from the perspective of the performance indicator. Moreover, in the realm of clinical evaluation using dose-volume histograms (DVH), MOEA/FMOP exhibits a better capability to generate high-quality solutions concurrently.

Class incremental medical segmentation (CIMS) aims to sequentially learn new classes while preserving knowledge of previously learned categories in the absence of old-class labels. Current methods suffer from performance degradation under class imbalance and require additional segmentation heads to accommodate new categories. Inspired by recent prototype learning that leverages prototypes to achieve robust recognition of new categories under limited-data regimes, we introduce a Prototype-dRIven class increMEntal (PRIME) method. PRIME replaces the incremental segmentation heads with prototypes to mitigate class imbalance, allowing new class learning with the simple addition of new prototypes. Based on prototype learning, PRIME further involves three tailored techniques. First, prototype structure alignment imposes structural constraints on inter-prototype relations to maintain consistent relative distances in the feature space, improving the model's ability to distinguish distinct classes. Second, pixel-wise contrastive loss term groups embeddings of similar samples while separating those of different classes, enhancing segmentation accuracy across all categories. Finally, the consensus-based prototype update mechanism refines the old prototypes during the learning of new classes, preventing performance degradation on the old classes. Extensive experiments on two public multi-organ segmentation datasets demonstrate that our approach significantly outperforms state-of-the-art methods, validating the effectiveness of the proposed PRIME.

Precisely segmenting the organs at risk (OARs) in computed tomography (CT) plays an important role in radiotherapy’s treatment planning, aiding in the protection of critical tissues during irradiation. Renowned deep convolutional neural networks (DCNNs) and prevailing transformer-based architectures are widely utilized to accomplish the segmentation task, showcasing advantages in capturing local and contextual characteristics. Graph convolutional networks (GCNs) are another specialized model designed for processing the nongrid dataset, e.g., citation relationship. The DCNNs and GCNs are considered as two distinct models applicable to the grid and nongrid datasets, respectively. Motivated by the recently developed dynamic-channel GCN (DCGCN) that attempts to leverage the graph structure to enhance the feature extracted by the DCNNs, this paper proposes a novel architecture termed adaptive sparse GCN (ASGCN) to mitigate the inherent limitations in DCGCN from the aspect of node’s representation and adjacency matrix’s construction. For the node’s representation, the global average pooling used in the DCGCN is replaced by the learning mechanism to accommodate the segmentation task. For the adjacency matrix, an adaptive regularization strategy is leveraged to penalize the coefficient in the adjacency matrix, resulting in a sparse one that can better exploit the relationships between nodes. Rigorous experiments on multiple OARs’ segmentation tasks of the head and neck demonstrate that the proposed ASGCN can effectively improve the segmentation accuracy. Comparison between the proposed method and other prevalent architectures further confirms the superiority of the ASGCN.

Deep neural networks (DNNs) have emerged as a prominent model in medical image segmentation, achieving remarkable advancements in clinical practice. Despite the promising results reported in the literature, the effectiveness of DNNs necessitates substantial quantities of high-quality annotated training data. During experiments, we observe a significant decline in the performance of DNNs on the test set when there exists disruption in the labels of the training dataset, revealing inherent limitations in the robustness of DNNs. In this paper, we find that the neural memory ordinary differential equation (nmODE), a recently proposed model based on ordinary differential equations (ODEs), not only addresses the robustness limitation but also enhances performance when trained by the clean training dataset. However, it is acknowledged that the ODE-based model tends to be less computationally efficient compared to the conventional discrete models due to the multiple function evaluations required by the ODE solver. Recognizing the efficiency limitation of the ODE-based model, we propose a novel approach called the nmODE-based knowledge distillation (nmODE-KD). The proposed method aims to transfer knowledge from the continuous nmODE to a discrete layer, simultaneously enhancing the model's robustness and efficiency. The core concept of nmODE-KD revolves around enforcing the discrete layer to mimic the continuous nmODE by minimizing the KL divergence between them. Experimental results on 18 organs-at-risk segmentation tasks demonstrate that nmODE-KD exhibits improved robustness compared to ODE-based models while also mitigating the efficiency limitation.

3-D radiotherapy is an effective treatment modality for breast cancer. In 3-D radiotherapy, delineation of the clinical target volume (CTV) is an essential step in the establishment of treatment plans. However, manual delineation is subjective and time consuming. In this study, we propose an automated segmentation model based on deep neural networks for the breast cancer CTV in planning computed tomography (CT). Our model is composed of three stages that work in a cascade manner, making it applicable to real-world scenarios. The first stage determines which slices contain CTVs, as not all CT slices include breast lesions. The second stage detects the region of the human body in an entire CT slice, eliminating boundary areas, which may have side effects for the segmentation of the CTV. The third stage delineates the CTV. To permit the network to focus on the breast mass in the slice, a novel dynamically strided convolution operation, which shows better performance than standard convolution, is proposed. To train and evaluate the model, a large dataset containing 455 cases and 50 425 CT slices is constructed. The proposed model achieves an average dice similarity coefficient (DSC) of 0.802 and 0.801 for right-0 and left-sided breast, respectively. Our method shows superior performance to that of previous state-of-the-art approaches.

A rational leader selection strategy can enhance a swarm to manage the convergence and diversity during the entire search process. In this article, a novel adaptive multiobjective particle swarm optimization (MOPSO) is proposed on the basis of an evolutionary state estimation mechanism, which is used to detect the evolutionary environment whether in exploitation or exploration state. During the search process, different types of leaders, such as a convergence global best solution (c-gBest) and several diversity global best solutions (d-gBests), are to be selected from the external archive for particles under different evolutionary environments. The c-gBest is selected for improving the convergence when the swarm is in an exploitation state, while the d-gBests are chosen for enhancing the diversity in an exploration state. Furthermore, a modified archive maintenance strategy based on some predefined reference points is adopted to maximize the diversity of the Pareto solutions in the external archive. The experimental results demonstrate that the proposed algorithm performs significantly better than the several state-of-the-art multiobjective PSO algorithms and multiobjective evolutionary algorithms on 31 benchmark functions in terms of convergence and diversity of those obtained approximate Pareto fronts.

Radiotherapy is an indispensable part of adjuvant therapy for cancer that improves local control, overall survival, and the opportunity for good quality of life. Organ delineation and dose plan design are the key steps in the treatment. Organ delineation controls the area of radiotherapy and dose planning controls its intensity. However, both tasks are time-consuming, exhausting, and subjective, and automated methods are desirable. Although automated methods have been studied, the previous studies either focus on organ segmentation or dose prediction, without considering them from a holistic perspective. In this paper, we treat organ segmentation and dose prediction as similar tasks, and propose an error correction framework to improve their performance based on the same mechanism. The proposed error correction framework consists of a prediction network and a calibration network. The biggest difference between our framework and previous studies is that the state-of-the-art networks can be used as a prediction network or calibration network, and then the performance can be improved by the error correction mechanism. To evaluate the framework, we conducted a series of experiments on dose prediction and organ segmentation. These experimental results show that the framework is superior to other state-of-the-art methods in both tasks.

Macular edema is a retinal disease that may cause visual loss, even blindness. Both Subretinal fluid (SRF) and Pigment epithelium detachment(PED) are significant characteristics to help diagnose this disease. Optical coherence tomography (OCT) is a recognized technology for scanning retinal tissue, due to its non-invasive and high-resolution properties. Classification of SRF and PED among OCT images is thus the main task for macular edema diagnosis. General classification methods are based on classical machine learning applying domain specific knowledge for designing hand-crafted features. We proposed a residual network model to classify SRF and PED features among OCT images. In order to achieve better performance, data augmentation is investigated to respond to the challenge of data shortage. And fine-tuning Residual network from pre-trained parameters is applied. Since the task is a multi-label problem where a single data may have two labels, we also explored the potential correlations between these two labels. The large OCT dataset for training and evaluating model is provided by a competition platform called "AI Challenger". Experiments on the large-scale AI-Challenger OCT dataset demonstrate the effectiveness of the proposed approach. As a result, we achieve accuracies of 99.01% and 98.65% for the classification of SRF and PED, respectively.

Retinopathy of Prematurity (ROP) is a retinal vasproliferative disorder disease principally observed in infants born prematurely with low birth weight. ROP is an important cause of childhood blindness. Although automatic or semi-automatic diagnosis of ROP has been conducted, most previous studies have focused on "plus" disease, which is indicated by abnormalities of retinal vasculature. Few studies have reported methods for identifying the "stage" of the ROP disease. Deep neural networks have achieved impressive results in many computer vision and medical image analysis problems, raising expectations that it might be a promising tool in the automatic diagnosis of ROP. In this paper, convolutional neural networks with a novel architecture are proposed to recognize the existence and severity of ROP disease per-examination. The severity of ROP is divided into mild and severe cases according to the disease progression. The proposed architecture consists of two sub-networks connected by a feature aggregate operator. The first sub-network is designed to extract high-level features from images of the fundus. These features from different images in an examination are fused by the aggregate operator, then used as the input for the second sub-network to predict its class. A large data set imaged by RetCam 3 is used to train and evaluate the model. The high classification accuracy in the experiment demonstrates the effectiveness of the proposed architecture for recognizing the ROP disease.