Chunliang Wang

Lung cancer remains the leading cause of cancer-related mortality worldwide (1), largely due to latestage diagnosis. Over the past decade, the implementation of low-dose CT screening has altered the clinical landscape by enabling earlier detection and improved survival outcomes (2,3). However, this shift has also introduced new challenges, including the interpretation of increasingly large imaging datasets, growing diagnostic complexity, the management of indeterminate pulmonary nodules, and the need for more individualized treatment strategies. In this context, artificial intelligence (AI) has emerged as a potentially important tool across different stages of lung cancer care. This special issue aimed to bring a range of studies that examine the role of AI in lung cancer screening and treatment, spanning early detection, diagnostic characterization, molecular inference, treatment decision-making, and prognostic modeling. Taken together, these contributions reflect a gradual transition from isolated algorithm development toward more clinically oriented and integrative approaches.A recurring theme in this issue is the expansion of lung cancer screening beyond conventional CTbased screening approaches. While CT remains central to current screening strategies (4), several studies explore multimodal approaches that integrate imaging with additional sources of clinical information. One direction involves the incorporation of serum biomarkers for non-invasive risk stratification and early disease assessment (Zhao et al.). Another direction is the use of alternative imaging modalities such as ultrasound, which may provide complementary diagnostic information in selected clinical scenarios (Wei et al.). These developments indicate that lung cancer screening may increasingly evolve toward a multimodal framework.Beyond nodule detection, several studies focus on improving diagnostic accuracy through quantitative imaging analysis. Radiomics-based models incorporating both intranodular and perinodular features demonstrate promising performance in differentiating pulmonary cryptococcosis from lung adenocarcinoma (Deng et al.), highlighting the potential value of capturing information from both the tumor and its surrounding environment. Similarly, models combining radiomic features and clinicalsemantics show promising performance in distinguishing histological subtypes of non-small cell lung cancer, particularly squamous cell carcinoma and adenocarcinoma (Li et al.). These approaches suggest that quantitative imaging analysis may provide complementary information beyond conventional visual assessment. By extracting high-dimensional features, AI-based methods may identify subtle imaging patterns that are difficult to appreciate visually, potentially supporting more accurate diagnostic characterization and guide further follow-up decisions.Another area of interest is the application of AI to link imaging findings with underlying tumor biology. Several studies report that radiomic and deep learning models can predict molecular characteristics, including gene expression patterns and mutation status (Li et al. and Luo et al.). The integration of radiomics with single-cell transcriptomic data also represents an emerging direction, enabling the linkage of imaging features with molecular phenotypes and metastasis-associated gene expression (Wu et al.). In addition, deep learning applied to histopathological images has been used to identify morphological patterns associated with treatment response, extending beyond conventional biomarkers such as PD-L1 expression (Peroz et al.). These findings show that AI-based approaches may contribute to noninvasive characterization of tumor biology, although further validation is required before clinical implementation.AI applications are also being explored in the context of treatment decision support. Models predicting immunotherapy response and immune-related adverse events illustrate the potential for integrating clinical and imaging features to support therapeutic planning (Cao et al. and Hou et al.). Such approaches may help to balance treatment efficacy and toxicity at an individual level. Furthermore, models trained on multidisciplinary tumor board decisions show that machine learning can approximate complex clinical decision processes (Pasello et al.). While these approaches are still at an early stage, they may have potential in supporting decision-making in settings with limited access to expert consensus.In addition to diagnosis and treatment selection, this special issue shows that AI is being applied to prognostic modeling and perioperative risk assessment. Studies predicting postoperative recurrence (Lan et al.) and pulmonary complications suggest that AI may support longitudinal patient management (Sha et al.). Early identification of high-risk patients could facilitate more tailored follow-up and intervention strategies. These applications indicate that AI may support multiple stages of the clinical workflow, extending beyond isolated tasks toward more continuous and longitudinal patient management.Methodological trends can be observed across the studies included in this issue. First, there is increasing use of multimodal data integration, combining imaging, clinical, and molecular information. Such approaches may provide more comprehensive characterization of tumor biology and support more individualized clinical assessment. Second, model interpretability is receiving growing attention, with approaches such as SHAP and Grad-CAM used to provide insights into model decision-making (5,6). Improved interpretability may facilitate clinical understanding and increase confidence in AIassisted decision-making. Third, increasing attention is being paid to external validation and generalizability, with multi-center datasets being used more frequently. Despite these advances, several challenges remain, including data heterogeneity, limited standardization of...

Background: Increased epicardial adipose tissue volume (EATV) is a potentially important risk marker for coronary artery disease (CAD) available from cardiac computed tomography (CT) images. Sex-differences and effects of age and body size on EATV have been insufficiently explored, and no reliable reference values exist. Consequently, EATV has yet to find its deserved use in clinical practice. Objectives: To define normal values by sex and age, the best normalization procedure for EATV to neutralize effects of body-size, explore the relationship between normalized EATV and cardiac risk, and propose a clinically meaningful cut-off. Methods: AI-based automated EATV data from the general population (n=25,155) and a clinical cohort (n=2,482) with suspected CAD was normalized to height, BSA and heart volumes. Correlation between EATV and EAT attenuation was tested with Spearman?s rank correlation and linear regression to find the optimal normalization. Normalized EATV was compared to high-risk by SCORE2 and obstructive CAD in the population cohort. A cut-off including 95% of cases with obstructive CAD was defined in the general population and tested in the clinical cohort. Results: EATV varied with sex and age across cohorts. Normalization of EATV to total heart volume (EATVh) was superior by all metrics and neutralized the effects of sex. High-risk by SCORE2 and the prevalence of obstructive CAD increased over quartiles of EATVh in the population cohort, and significantly higher EATVh was seen with obstructive CAD in both cohorts. A cut-off of 0.1 in EATVh had a negative predictive value for obstructive CAD of 97.1% in the general population and 88.9% in the clinical cohort. Conclusions: EATV varies considerably with sex, age and body size. Normalization to heart volume outperformed other procedures, and EATVh is a useful marker of obstructive CAD in both the general population and symptomatic patients.

Background: Angiography‐derived index of microcirculatory resistance (IMR) has emerged as a feasible alternative to wire‐based IMR, due to the elimination of specialized wire and simplifying of the procedure. However, the differences in diagnostic accuracy between hyperemic and resting angiography with this technology remain underexplored. Methods: This prospective, multicenter study was conducted to evaluate the feasibility and diagnostic accuracy of x‐ray angiography–derived IMR (xaIMR) in patients with chronic coronary syndrome using wire‐based functional indices as the reference. Angiography‐derived functional indices were calculated using a computational fluid dynamics model. The xaIMR indices were derived from hyperemic and resting angiography, respectively. Results: A total of 139 vessels were included in the analysis of the present study. Both hyperemic xaIMR ( r =0.82; P r =0.25; P =0.003) exhibited positive linear correlations with wire‐based IMR. Notably, hyperemic xaIMR was associated with a higher diagnostic yield (area under the receiver operating characteristic curve) than resting xaIMR (area under the receiver operating characteristics curve, 0.95 versus 0.72; P <0.001). Significant differences in diagnostic yield between these modalities were observed in vessels with compromised coronary flow reserve (<2.0). In contrast, such differences were not observed in vessels with normal coronary flow reserve (≥2.0). Conclusions: xaIMR, especially derived from hyperemic angiography, is a reliable, noninvasive alternative to wire‐based IMR for assessing coronary microcirculatory resistance in chronic coronary syndrome. However, resting angiography‐derived xaIMR may have limited diagnostic accuracy in patients with reduced coronary flow reserve. Further studies are warranted to validate these findings in broader populations.

Lung lobe segmentation plays a significant role in the evaluation and treatment of various lung pathologies. This anatomical information becomes crucial for assessing the distribution and extent of pulmonary diseases.^1,2 Additionally, this information aids in optimizing treatment planning for such pathologies by focusing on the affected lobes.³ Many lung diseases primarily impact specific lung lobes,^4,5 underscoring the relevance of lung lobe segmentation in diagnosing and evaluating these conditions.

Intracranial hemorrhage (ICH) is a common finding in traumatic brain injury (TBI) and computed tomography (CT) is considered the gold standard for diagnosis. Automated detection of ICH provides clinical value in diagnostics and in the ability to feed robust quantification measures into future prediction models. Several studies have explored ICH detection and segmentation but the research process is somewhat hindered due to a lack of open large and labeled datasets, making validation and comparison almost impossible. The complexity of the task is further challenged by the heterogeneity of ICH patterns, requiring a large number of labeled data to train robust and reliable models. Consequently, due to the labeling cost, there is a need for label-efficient algorithms that can exploit easily available unlabeled or weakly-labeled data. Our aims for this study were to evaluate whether transfer learning can improve ICH segmentation performance and to compare a variety of transfer learning approaches that harness unlabeled and weakly-labeled data. Three self-supervised and three weakly-supervised transfer learning approaches were explored. To be used in our comparisons, we also manually labeled a dataset of 51 CT scans. We demonstrate that transfer learning improves ICH segmentation performance on both datasets. Unlike most studies on ICH segmentation our work relies exclusively on publicly available datasets, allowing for easy comparison of performances in future studies. To further promote comparison between studies, we also present a new public dataset of ICH-labeled CT scans, Seq-CQ500.

Head and neck surgery is a fine surgical procedure with a complex anatomical space, difficult operation and high risk. Medical image computing (MIC) that enables accurate and reliable preoperative planning is often needed to reduce the operational difficulty of surgery and to improve patient survival. At present, artificial intelligence, especially deep learning, has become an intense focus of research in MIC. In this study, the application of deep learning-based MIC in head and neck surgery is reviewed. Relevant literature was retrieved on the Web of Science database from January 2015 to May 2022, and some papers were selected for review from mainstream journals and conferences, such as IEEE Transactions on Medical Imaging, Medical Image Analysis, Physics in Medicine and Biology, Medical Physics, MICCAI, etc. Among them, 65 references are on automatic segmentation, 15 references on automatic landmark detection, and eight references on automatic registration. In the elaboration of the review, first, an overview of deep learning in MIC is presented. Then, the application of deep learning methods is systematically summarized according to the clinical needs, and generalized into segmentation, landmark detection and registration of head and neck medical images. In segmentation, it is mainly focused on the automatic segmentation of high-risk organs, head and neck tumors, skull structure and teeth, including the analysis of their advantages, differences and shortcomings. In landmark detection, the focus is mainly on the introduction of landmark detection in cephalometric and craniomaxillofacial images, and the analysis of their advantages and disadvantages. In registration, deep learning networks for multimodal image registration of the head and neck are presented. Finally, their shortcomings and future development directions are systematically discussed. The study aims to serve as a reference and guidance for researchers, engineers or doctors engaged in medical image analysis of head and neck surgery.

Brain MRI segmentation results should always undergo a quality control (QC) process, since automatic segmentation tools can be prone to errors. In this work, we propose two deep learning-based architectures for performing QC automatically. First, we used generative adversarial networks for creating error maps that highlight the locations of segmentation errors. Subsequently, a 3D convolutional neural network was implemented to predict segmentation quality. The present pipeline was shown to achieve promising results and, in particular, high sensitivity in both tasks.