Applied Intelligence最新文献_第4页

Diffusion-inspired prototypical network for few-shot fine-grained classification 稀疏细粒度分类的扩散启发原型网络

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-02 DOI: 10.1007/s10489-026-07509-2

Jia Min Lim, Kian Ming Lim, Pey Yun Goh, Chin Poo Lee

{"title":"Diffusion-inspired prototypical network for few-shot fine-grained classification","authors":"Jia Min Lim, Kian Ming Lim, Pey Yun Goh, Chin Poo Lee","doi":"10.1007/s10489-026-07509-2","DOIUrl":"10.1007/s10489-026-07509-2","url":null,"abstract":"<div><p>Distinguishing minute visual differences under conditions of extreme data scarcity remains a critical challenge in few-shot fine-grained image classification. When relying on limited support samples, existing methods frequently overfit to spurious background details rather than learning robust semantic representations, limiting their generalization ability. To address this issue, we propose Diffusion-inspired Prototypical Network (DiffProNet), a framework that incorporates a self-supervised feature-space denoising objective into the few-shot learning pipeline. Rather than performing image generation, the proposed method injects Gaussian noise into the support feature representations and employs a Feature Denoising Head to predict the injected noise through a self-supervised auxiliary objective. This denoising objective encourages the encoder to learn robust and discriminative feature representations, thereby regularizing the latent feature space and mitigating overfitting to irrelevant visual details. Extensive experiments on the CUB-200-2011, Stanford Dogs, and Stanford Cars benchmarks demonstrate the effectiveness of the proposed approach, yielding competitive performance under both 1-shot and 5-shot settings and validating the effectiveness of feature-space denoising for few-shot fine-grained image classification.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132395","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Solving risk-aware HTN planning problems: algorithmic, formal, and empirical results 解决有风险意识的HTN规划问题：算法、形式和实证结果

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-02 DOI: 10.1007/s10489-026-07450-4

Ebaa Alnazer, Ilche Georgievski, Marco Aiello

{"title":"Solving risk-aware HTN planning problems: algorithmic, formal, and empirical results","authors":"Ebaa Alnazer, Ilche Georgievski, Marco Aiello","doi":"10.1007/s10489-026-07450-4","DOIUrl":"10.1007/s10489-026-07450-4","url":null,"abstract":"<div><p>Actual, realistic domains are characterised by inherent uncertainty, compelling the development of planning strategies that enable the embracing of risks, where performing actions incurs costs, such as fuel and time, that cannot be predefined as constants during planning. In such domains, an agent’s perspective of desired plans is widened to not only accept plans with an average expected cost but also plans that comply with its attitude towards risk. For instance, in autonomous vehicle navigation, unforeseen obstacles and changing traffic conditions introduce significant risks that necessitate adapting driving tasks and route choices to comply with a particular risk attitude, ranging from extreme risk aversion to extreme risk seeking. Risk-aware Hierarchical Task Network (HTN) planning, a general framework that extends classical HTN planning with capabilities for embracing risk by leveraging established concepts from decision theory, is specifically developed to address planning problems in real-world domains. Here, we develop an approach that solves a subset of planning problems defined by that general framework, thereby surpassing the expressiveness of current approaches by allowing HTN planning agents to compute plans tailored to a particular risk attitude. We empirically evaluate the approach on four risk-involving HTN planning domains, perform an ablation study isolating the individual contributions of the approach, and compare the approach against four state-of-the-art HTN planners, including cost-unaware and cost-optimal. The results demonstrate the feasibility and expressiveness of the approach, showing that it is on par with established state-of-the-art HTN planners while producing plans that are meaningfully sensitive to risk. We also discuss open issues raised by the proposed approach, including their adaptability to solve other risk-aware HTN planning problems.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07450-4.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132066","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Probabilistic periodic frequent pattern mining in uncertain databases 不确定数据库中的概率周期频繁模式挖掘

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-02 DOI: 10.1007/s10489-026-07388-7

Ilyes Hamimid, Farid Nouioua, Sara Boutouhami, Philippe Fournier-Viger

{"title":"Probabilistic periodic frequent pattern mining in uncertain databases","authors":"Ilyes Hamimid, Farid Nouioua, Sara Boutouhami, Philippe Fournier-Viger","doi":"10.1007/s10489-026-07388-7","DOIUrl":"10.1007/s10489-026-07388-7","url":null,"abstract":"<div><p>In recent years, the detection of periodic frequent patterns in transactional datasets has attracted considerable attention due to their applicability across various domains. However, while existing studies have primarily focused on deterministic data, existing approaches for uncertain data either rely on expected support models that lack probabilistic guarantees, or employ possible-world semantics with strong simplifying assumptions (e.g., transaction-level uncertainty) and high computational overhead. Prior work such as SPFIM introduces possible-world notions for uncertain periodic pattern mining but is restricted to transaction-level uncertainty and does not provide a fully specified algorithmic procedure for computing probabilistic periodicity during mining, while expected-support-based methods such as UPFP-Growth++ achieve runtime efficiency but do not quantify the probability that a pattern is periodic. These limitations undermine the reliability of extracted patterns, as neither approach adequately quantifies the confidence in a pattern’s periodic behavior. To address these challenges, this study establishes a formal framework for probabilistic periodic pattern mining under possible-world semantics that explicitly models fine-grained item-level existential probabilities under the standard independence assumption used in uncertain itemset mining. Within this framework, a novel pattern-growth algorithm named PPFP-Growth is proposed for mining probabilistic periodic frequent patterns in uncertain transaction databases. To overcome the challenge of efficiently processing multiple possible worlds, the algorithm integrates a novel decomposition method. Empirical evaluations on five benchmark datasets under controlled item-level uncertainty settings, including a comparison with UPFP-Growth++, show that the proposed algorithm is practical for extracting probabilistic periodic frequent patterns while quantifying the additional computational cost required to provide probabilistic guarantees on periodicity.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132065","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Condition-Controlled optimal transport for cellular perturbation response prediction 细胞扰动响应预测的条件控制最优运输

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-02 DOI: 10.1007/s10489-026-07521-6

Jiaming Liu, Qi Tan, Pei Yang

{"title":"Condition-Controlled optimal transport for cellular perturbation response prediction","authors":"Jiaming Liu, Qi Tan, Pei Yang","doi":"10.1007/s10489-026-07521-6","DOIUrl":"10.1007/s10489-026-07521-6","url":null,"abstract":"<div><p>Recent advances in single-cell sequencing have significantly enhanced our understanding of cellular heterogeneity. However, predicting cellular responses to perturbations remains a grand challenge. Several critical limitations impede progress: the unpaired nature of perturbation data as single-cell sequencing techniques necessitate cellular disruption, and the need for generalization to novel therapeutic compounds. To this end, we propose Condition-Controlled Optimal Transport (CCOT) to model cellular perturbation responses. CCOT addresses the challenges by integrating three key innovations, the condition-controlled mechanism, cross-drug generalization and alignment with biological priors, into a principled framework. Specifically, CCOT introduces a novel condition-controlled mechanism to bridge conditional and unconditional models based on optimal transport. Leveraging partially input convex neural networks to learn condition-controlled transport maps between unperturbed and perturbed cellular distributions, CCOT enables in-distribution prediction and out-of-distribution generalization to novel compounds. Comprehensive evaluations on the SciPlex3 dataset demonstrate that CCOT significantly outperforms state-of-the-art methods in capturing complex and nonlinear cellular responses. Its superiority is more evident in capturing higher-order statistical properties and gene-specific responses for differentially expressed genes. We believe these findings offer valuable insights and a powerful tool for advancing drug discovery and understanding cellular mechanisms.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132394","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multi-view self-supervised learning for label-flipping robustness in tabular data: a comparative study 表格数据中标签翻转鲁棒性的多视图自监督学习：比较研究

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-01 DOI: 10.1007/s10489-026-07452-2

Jinhyeok Jang, Sohee Park, SeungSoo Nam, Daeseon Choi

{"title":"Multi-view self-supervised learning for label-flipping robustness in tabular data: a comparative study","authors":"Jinhyeok Jang, Sohee Park, SeungSoo Nam, Daeseon Choi","doi":"10.1007/s10489-026-07452-2","DOIUrl":"10.1007/s10489-026-07452-2","url":null,"abstract":"<div><p>Tabular machine learning models are widely used in high-stakes domains such as finance, healthcare, and network intrusion detection, where their reliability depends heavily on accurate training labels. However, malicious or accidental label flipping can corrupt training data and degrade downstream model performance. Recent attacks further target decision-boundary or vulnerable local regions rather than flipping labels randomly. This paper presents a BYOL-based multi-view KNN sanitization approach for identifying suspicious label-flipped samples in tabular data. The method learns label-independent representations and examines KNN label-disagreement patterns across original and perturbed SSL embedding views. We evaluate the method on six cross-domain tabular benchmarks under both large-scale and small-scale label-flipping attacks, and compare it with existing sanitization baselines and a SCARF-based representation baseline. The results show that multi-view neighborhood probing improves suspicious-label detection and provides a practical F1-oriented trade-off, while downstream recovery remains attack- and target-model-dependent. These findings provide actionable insights for improving the robustness and label integrity of tabular machine learning systems against label-flipping attacks.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07452-2.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132201","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Auditing weakly supervised cryptocurrency risk-state learning: when label fit, stability, and semantics disagree 审计弱监督的加密货币风险状态学习：当标签匹配、稳定性和语义不一致时

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-01 DOI: 10.1007/s10489-026-07467-9

Junwen Lou, Fukai Zhang, Shuo Tian

{"title":"Auditing weakly supervised cryptocurrency risk-state learning: when label fit, stability, and semantics disagree","authors":"Junwen Lou, Fukai Zhang, Shuo Tian","doi":"10.1007/s10489-026-07467-9","DOIUrl":"10.1007/s10489-026-07467-9","url":null,"abstract":"<div><p>Heuristic proxy labels are easy to construct, but what they transfer to learned market-state representations remains unclear. This paper proposes an audit protocol for weakly supervised cryptocurrency risk-state learning rather than another forecasting architecture. Using Binance spot 1-hour data for five assets, the protocol evaluates candidate state learners along three separated axes: proxy-label fit, temporal stability, and ex-post risk semantics. Across the tested assets, lightweight baselines, and BTCUSDT temporal-backbone stress tests, state paths align more consistently with future risk stratification than with return direction, but this result is conditional on the specific weak-label design used here. In particular, the recovered risk layering is materially shaped by the downside-risk information encoded through the <span>max_drawdown_72</span> anchor in both the feature set and the label rule. As a result, the model with the best proxy-label fit need not yield the most credible state path, and conventional multi-criteria selectors such as weighted scoring and TOPSIS do not reliably detect that gap once stronger label-fitting candidates enter the pool. Under the tested settings, added semantic regularization provides no measurable gain, and lightweight Transformer- and PatchTST-style references do not overturn the primary trade-off. The paper also identifies a boundary case: XRP proxy labels are already misaligned with the target risk semantics before model training, and the proposed Semantic Orientation Ratio detects that misalignment at the label-audit stage. Overall, the contribution is a protocol-level audit view of weak supervision in which label fidelity, path stability, semantic validity, and label-design dependence can disagree.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132255","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A dual conditional modulation graph attention network for temporal link prediction 一种用于时间链路预测的双条件调制图注意网络

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-01 DOI: 10.1007/s10489-026-07476-8

Weijian Zhong, Xian Mu, Dagang Li, Zhongwei Huang, Wei Ren

{"title":"A dual conditional modulation graph attention network for temporal link prediction","authors":"Weijian Zhong, Xian Mu, Dagang Li, Zhongwei Huang, Wei Ren","doi":"10.1007/s10489-026-07476-8","DOIUrl":"10.1007/s10489-026-07476-8","url":null,"abstract":"<div><p>Temporal link prediction plays a pivotal role in numerous real-world applications, such as communication network forecasting, social network evolution analysis, and financial fraud detection. In these scenarios, accurately predicting future interactions is not only valuable but essential for ensuring timely decision-making, personalized services, and system stability. However, effectively performing temporal link prediction in dynamic networks remains challenging due to the need to jointly capture complex, evolving dependencies across multiple information dimensions—namely attribute features, structural patterns, and temporal dynamics. To address this issue, we propose DCM-Net, a novel Dual Conditional Modulation Graph Attention Network. Our framework introduces a sophisticated dual-level control mechanism that dynamically optimizes both the input features and the learning process of the graph attention network. Specifically, we first introduce an Adaptive Feature Modulation module, which acts at the input-level by purifying node attributes to enhance task-relevant signals. Subsequently, a Temporal Parameter Modulation module, acts at the process-level. It leverages a temporal model to dynamically evolve the core parameters of the GAT(Graph Attention Networks) encoder itself, thereby making the structural aggregation process adaptive to the network’s temporal evolution. Experimental results on six real-world dynamic networks show that DCM-Net achieves the best average performance across ROC-AUC, MAP, and PR-AUC, with statistically significant improvements in several comparisons.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132083","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

ANNE: adaptive neuron network expansion for 2D medical image segmentation 基于自适应神经元网络扩展的二维医学图像分割

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-01 DOI: 10.1007/s10489-026-07503-8

Abel A. Reyes-Angulo, Sidike Paheding

{"title":"ANNE: adaptive neuron network expansion for 2D medical image segmentation","authors":"Abel A. Reyes-Angulo, Sidike Paheding","doi":"10.1007/s10489-026-07503-8","DOIUrl":"10.1007/s10489-026-07503-8","url":null,"abstract":"<div><p>Accurate 2D medical image segmentation requires both fine-scale boundary modeling and long-range contextual reasoning, yet architectures that provide both capabilities are often computationally expensive. We present ANNE, a compact encoder–decoder that combines Mamba-based state-space blocks for efficient global-context modeling with Adaptive Progressive Expansion Neurons (A-PEN) for nonlinear local feature refinement. A-PEN extends prior progressively expanded neuron formulations by introducing learnable normalized gates over a fixed bank of polynomial expansion terms, together with trainable expansion coefficients and a parametric activation. The normalized gates provide a coupled parameterization of the candidate expansion branches rather than a uniquely identifiable measure of individual term importance. Remez approximation is used as approximation-theoretic motivation for compact polynomial representations; ANNE does not execute the classical Remez exchange algorithm or claim a minimax approximation guarantee. We evaluate ANNE on eight public 2D datasets covering dermoscopic lesion, colonoscopic polyp, and retinal-vessel segmentation. Across these benchmarks, ANNE provides competitive overlap and boundary accuracy while using approximately 2.77 million trainable parameters. Component ablations indicate that A-PEN and Mamba make complementary contributions, and resolution profiling identifies a favorable accuracy–cost operating point at <span>(varvec{256times 256})</span> pixels. To ensure transparent interpretation, controlled reproductions are distinguished from literature-reported comparisons. The results support ANNE as a parameter-efficient segmentation architecture within the evaluated 2D benchmark setting, while external-domain, volumetric, and clinical validation remain important directions for future work.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10489-026-07503-8.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132199","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Dynamic embedding and implicit relation-aware self-attention for TKGC TKGC的动态嵌入和隐式关系感知自关注

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-01 DOI: 10.1007/s10489-026-07481-x

Shuang Liu, Xiaohui Sun, Peng Chen, Simon Kolmanič

{"title":"Dynamic embedding and implicit relation-aware self-attention for TKGC","authors":"Shuang Liu, Xiaohui Sun, Peng Chen, Simon Kolmanič","doi":"10.1007/s10489-026-07481-x","DOIUrl":"10.1007/s10489-026-07481-x","url":null,"abstract":"<div><p>Temporal Knowledge Graph Completion (TKGC) is a fundamental reasoning task for dynamically evolving knowledge, aiming to infer missing facts in time-varying relational data and support time-sensitive decision-making. It is particularly critical in extrapolation scenarios, including event forecasting and temporal question answering. However, structural sparsity and inadequate modeling of multi-dimensional temporal entity evolution still severely limit existing TKGC methods. Most existing approaches fail to simultaneously capture explicit structural connections and implicit semantic associations, and lack effective modeling of entity temporal evolution, resulting in inferior performance in sparse and long-term reasoning scenarios. To mitigate these issues, we propose DIRA: Dynamic embedding and Implicit Relation-aware Self-attention model for TKGC. DIRA employs multi-component dynamic embedding to explicitly model entity trend, periodic, and static properties. It enhances structural and semantic representations via relation-aware graph convolution and adaptive implicit semantic completion graphs, and uses gating mechanisms and temporal self-attention to fuse multi-source information dynamically. Extensive experiments on four real-world datasets (ICEWS14, ICEWS18, ICEWS05-15, GDELT) show that DIRA consistently outperforms state-of-the-art baselines in MRR and Hits@1/3/10. The results validate its effectiveness and robustness for extrapolation reasoning , alleviating structural sparsity and improving temporal evolution modeling. This work demonstrates that combining explicit dynamic embedding with implicit semantic relation modeling is a promising direction for TKGC in sparse and extrapolation scenarios.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132200","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A needle path planning method for liver tumor radiofrequency ablation based on medical image segmentation deep neural networks 基于医学图像分割深度神经网络的肝脏肿瘤射频消融针路规划方法

IF 3.5 2区计算机科学

Applied Intelligence Pub Date : 2026-10-01 DOI: 10.1007/s10489-026-07323-w

Dan Liu, Tianjiao Duo

{"title":"A needle path planning method for liver tumor radiofrequency ablation based on medical image segmentation deep neural networks","authors":"Dan Liu, Tianjiao Duo","doi":"10.1007/s10489-026-07323-w","DOIUrl":"10.1007/s10489-026-07323-w","url":null,"abstract":"<div><p>Accurate preoperative needle path planning is vital for liver tumor Radiofrequency Ablation (RFA), yet conventional planning remains subjective and inefficient. This study proposes an automated planning framework centered on a lightweight and optimized medical image segmentation network. We introduce a novel Bezier curve-based stylistic feature extraction and style transfer strategy to mitigate domain gaps across heterogeneous datasets. Furthermore, multiple lightweight convolutions and attention mechanisms are integrated to achieve precise segmentation of the liver and critical risk structures. Specifically, clinical requirements—including needle length, incident angle, and safety margins—are transformed into multi-dimensional grayscale constraint maps. A Pareto optimization algorithm, integrated with a weighted product approach, is then employed to resolve the multi-objective planning problem. Experimental results on the LiTS dataset demonstrate that our model achieves a Dice score of 98.13%, comparable to state-of-the-art (SOTA) models while reducing parameters by 75%. Moreover, the proposed framework maintains a 86.7% success rate in trajectory planning under complex anatomical scenarios. This approach significantly enhances the objectivity and efficiency of RFA planning, providing robust technical support for intelligent and precision liver intervention.</p></div>","PeriodicalId":8041,"journal":{"name":"Applied Intelligence","volume":"56 15","pages":""},"PeriodicalIF":3.5,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132073","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0