Machine Learning Algorithms Research Articles

Machine learning-based prediction of postoperative mortality risk after abdominal surgery

BACKGROUND Preoperative risk assessments are vital for identifying patients at high risk of postoperative mortality. However, traditional scoring systems can be time consuming. We hypothesized that the use of machine learning models would enable rapid and accurate risk assessments to be performed. AIM To assess the potential of machine learning algorithms to develop predictive models of mortality risk after abdominal surgery. METHODS This retrospective study included 230 individuals who underwent abdominal surgery at the Seventh People’s Hospital of Shanghai University of Traditional Chinese Medicine between January 2023 and December 2023. Demographic and surgery-related data were collected and used to develop nomogram, decision-tree, random-forest, gradient-boosting, support vector machine, and naïve Bayesian models to predict 30-day mortality risk after abdominal surgery. Models were assessed using receiver operating characteristic curves and compared using the DeLong test. RESULTS Of the 230 included patients, 52 died and 178 survived. Models were developed using the training cohort (n = 161) and assessed using the validation cohort (n = 68). The areas under the receiver operating characteristic curves for the nomogram, decision-tree, random-forest, gradient-boosting tree, support vector machine, and naïve Bayesian models were 0.908 [95% confidence interval (CI): 0.824-0.992], 0.874 (95%CI: 0.785-0.963), 0.928 (95%CI: 0.869-0.987), 0.907 (95%CI: 0.837-0.976), 0.983 (95%CI: 0.959-1.000), and 0.807 (95%CI: 0.702-0.911), respectively. CONCLUSION Nomogram, random-forest, gradient-boosting tree, and support vector machine models all demonstrate strong performances for the prediction of postoperative mortality and can be selected based on the clinical circumstances.

Machine learning-assisted investigation on nonlinear vibration analysis of bio-inspired auxetic tubes

PurposeAuxetic tubular structures with negative Poisson’s ratios have gained significant attention in biomedical applications, particularly in vascular and esophageal stents, due to their potential to reduce embolism risks. This study aims to investigate the nonlinear vibration characteristics of such structures and develop accurate predictive models using machine learning (ML) techniques.Design/methodology/approachThe governing equations of auxetic tubes are derived using Hamilton’s principle and von-Kármán’s nonlinear assumptions, while Malek-Gibson relations determine their effective mechanical properties. The mechanical behavior of polylactic acid (PLA) is experimentally analyzed through tensile testing and digital image correlation (DIC) with additional insights from scanning electron microscopy. The nonlinear vibration equations are solved via the Ritz method, and vibrational behavior is assessed using the direct displacement control approach. Predictive modeling is performed using six ML algorithms – CatBoost, decision tree, random forest, gradient boosting tree, extreme gradient boosting (XGBoost) and support vector regression (SVR) – along with an artificial neural network (ANN). Response surface methodology is employed to optimize the effects of edge supports, radius ratios and auxetic cell geometry on vibrational behavior.FindingsThe results demonstrate a strong agreement between ML/ANN predictions and the analytical Ritz method, confirming the reliability of the developed models. The analysis reveals that variations in edge supports, radius ratios and auxetic cell geometry significantly influence the vibrational response of the structures. The optimized configurations enhance the structural performance, making these auxetic tubular metastructures highly suitable for biomedical applications.Originality/valueThis study uniquely integrates analytical modeling, experimental analysis and ML-based predictive modeling to comprehensively assess and optimize the vibrational behavior of auxetic tubular metastructures. The findings provide valuable insights for the design of next-generation auxetic stents, improving their mechanical performance and expanding their potential biomedical applications.

Composite Structural Health Prediction Using Vibroacoustic Responses via Machine Learning Techniques

This research adopted five different machine learning (ML) algorithms to identify damage in the laminated structure utilizing the vibroacoustic responses obtained via coupled finite element (FE)–boundary element (BE) solutions. Initially, the composite structural model is derived through higher-order midplane kinematics and solved to compute the fluid–structure interactions in terms of vibroacoustic values via in-house computer code. Furthermore, the exactness and effectiveness of the suggested model have been verified with experimental responses of the intact and damaged shell structure. Subsequently, the sound pressure level data have been extracted for different excitation frequencies numerically using various input parameters via the established model. First, the said ML models are trained using 1500 extracted data sets (using the customized MATLAB code) and predicted the damage using 303 datasets by setting the numerals, i.e., ‘0’ and ‘1’. A total of 1803 datasets (crack: 902 datasets; intact: 901 datasets) have been utilized for the current analysis. The current predicted responses indicate that the Random Forest Classifier is capable of providing higher accuracy (ranges between 81% and 91%) in comparison to all types of ML algorithms (Random Forest Classifier, Logistics Regression, Linear Support Vector Classified, Kernel Support Vector Machines and Artificial Neural Network) adopted. The proposed method detects and forecasts defects in the composite structures early and helps prevent adverse effects.

A Comparison Between Allied Ordnance Procedure‐48 and Various Machine Learning Models

ABSTRACTThis study examines the stabilizer depletion data of seven distinct gun propellants (single base and double base, stabilized with DPA or Arkadite II) using the allied ordnance procedure (AOP)‐48 kinetic approach and three machine learning algorithms: random forest, extreme gradient boosting, and neural network. The efficacy of the various methodologies is evaluated in relation to the quantity of training data. The AOP‐48 kinetic model demonstrates optimal performance when trained on sufficiently sized datasets, accurately predicting the stabilizer content with a mean absolute error of 0.03%. The mean absolute errors achieved by the machine learning algorithms were between 0.06% and 0.15% for stabilizer content. Nevertheless, the efficacy of machine learning models can be enhanced by incorporating the propellant composition into their architecture, thereby reducing the mean absolute error to a range of 0.05%–0.075% for the stabilizer content. The impact of varying training and testing data partitions has been subjected to comprehensive analysis, and the requisite quantity of data points for developing a model yielding accurate predictions (below 0.05% error of stabilizer concentration) has been determined to approximately 30 data points per formulation, while approximately 15 points are sufficient for the AOP‐48 procedure.

Forecasting US GDP in the Post-pandemic Age: An Evaluation of AR Model and Machine Learning Algorithms

The AR (1) model outperforms the random forest, KNN, gradient boosting, and Xgboost algorithms by yielding a lower MAE in the prediction result. The reason behind may be attributed to AR models ability to utilize data after a sudden shock. Since AR model predict one years GDP based on the previous years GDP, after a sudden shock, the effect of the shock will be captured when predicting the GDP of the year after the shock. But the machine learning algorithms evaluated in this research are trained using data in a fixed time frame. They are not able to utilize data in the testing data set and adjust accordingly. By evaluating the performance of the models, this research reveals the importance of a models ability to handle information of a sudden shock in GDP forecasting and that AR model possess such ability while random forest, KNN, gradient boosting, and Xgboost do not.

Metagenomic Microbial Signatures for Noninvasive Detection of Pancreatic Cancer

Background/Objectives: Pancreatic ductal adenocarcinoma (PDAC) is a highly aggressive malignancy with poor early detection rates owing to the limited sensitivity and specificity of the current biomarker CA19-9. Gut microbiota dysbiosis plays a key role in PDAC pathogenesis. This study aimed to evaluate the noninvasive approach we developed, combining metagenome-derived microbial signatures with CA19-9, to improve PDAC detection. Methods: This study included 50 treatment-naïve patients with PDAC and their matched controls. Fecal samples were analyzed using shotgun metagenomic sequencing. Machine learning algorithms were used to develop and validate a diagnostic panel integrating microbial signatures and CA19-9 levels. Subgroup analyses were used to confirm the robustness of the microbial markers. Results: The combined models at both species and genus levels effectively distinguished patients with PDAC from healthy individuals, and their strong diagnostic efficacy and accuracy were demonstrated through rigorous validation. Conclusions: In conclusion, the combination of gut microbiome profiling and CA19-9 improves PDAC detection accuracy compared to the use of CA19-9 alone, showing promise for early and noninvasive diagnosis.

Abstract 2419: Impact of immune contexture on immunotherapy response in NSCLC: insights from multiplex IHC and machine learning-based phenotyping

Abstract Background: The relationship between immune phenotypes (infiltrated, excluded, and desert) and response to immunotherapy in non-small cell lung cancer (NSCLC) is an area of active investigation. Understanding the tumor immune contexture can indeed provide valuable insights into the tumor microenvironment (TME) and immune dynamics, potentially guiding personalized therapeutic strategies. Methods: We analyzed a discovery cohort of 148 NSCLC patients using the CD8/PanCK multiplex immunohistochemistry (mIHC) panel to define the immune phenotype. Tumors with <2% of CD8+ cells were classified as "desert." Tumors with >2% CD8+ cells within the tumor area were categorized as "infiltrated, " while those with >2% CD8+ cells restricted to the periphery were labeled "excluded." Correlations between immune phenotypes and immunotherapy responses (PFS, OS and response rates) were assessed. Validation was performed on an independent cohort of 117 patients. Results: In the discovery cohort, tumors with an "infiltrated" phenotype demonstrated significantly better responses to immunotherapy compared to "excluded" and "desert" tumors (p = 0.009 “infiltrated” vs “desert”, p=0.045 “infiltrated” vs “excluded”). Immune exclusion was associated with lack of response. These findings were confirmed in the validation cohort, reinforcing the prognostic value of immune phenotyping in NSCLC. Additionally, tertiary lymphoid structures (TLS) and macrophage infiltration, known predictors of response to immunotherapy in NSCLC, were analyzed using an ISO-compliant multiplex immunohistofluorescence (mIHF) panel. Machine learning algorithms were employed to automate immune phenotyping and produce rapid summaries of TME characteristics. The multiplex IHF panel reliably assessed immune phenotypes, TLS presence, and macrophage infiltration with high reproducibility. Automated phenotyping using machine learning enabled rapid and accurate characterization of TME features, facilitating the integration of immune phenotyping into clinical workflows. Conclusions: Tumor immune contexture significantly influences the outcomes of immunotherapy in NSCLC. "Infiltrated" tumors are associated with favorable responses, while immune exclusion predicts resistance. Our ISO-certified multiplex IHF panel, combined with machine learning automation, offers a robust and scalable approach to immune phenotyping, supporting its implementation in precision oncology. Keywords: NSCLC, immune phenotyping, multiplex IHC, immune exclusion, infiltrated tumors, tertiary lymphoid structures, macrophage infiltration, machine learning, immunotherapy response. Citation Format: Jean-Philippe Guegan, Imane Nafia, Florent Petraud, Lucile Vanhersecke, Christophe Rey, Alban Bessede, Antoine Italiano. Impact of immune contexture on immunotherapy response in NSCLC: insights from multiplex IHC and machine learning-based phenotyping [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2419.

Abstract 2400: Utilizing machine learning to predict clinical staging with head and neck cancer gene expression datasets

Abstract Introduction: Head and neck cancer is a deadly disease with relatively stagnant cure rates. Predictive oncology might allow treatment stratification for surgery, radiation, and chemotherapy combinations based on both stage and other risks. As more high dimensional data from genomic expression datasets become publicly available, machine learning algorithms could be used for staging/treatment refinement. In the present study, we applied machine learning algorithms to our previously published smaller U133A Affymetrix gene expression dataset and the NCI Cancer Genome Atlas (TCGA). Methods: Models were trained to classify T and N staging data for TCGA and Affymetrix chip datasets, by early (T1/2) versus advanced T stage (T3+), and for N0 versus N1+, respectively, for TCGA and Affymetrix data separately. TCGA data was split according to HPV status into two datasets. Models to predict early vs. advanced stage were trained and tested on randomly (proportional with respect to staging) split data: large dataset (HPV- T and HPV- N, n>=350) models were trained on 80% and tested on the unseen 20% samples, while for the remainder (n<=50), data were split 70%-30% for sufficient testing. Scikit-learn models utilized were: Bagged (aggregated) Support Vector Classifiers (BSVCs), Random Forest (RF), Gradient Boosting (GB), and Linear Regression (LR). LR was added unaltered as a baseline prediction, while the other three models were subjected to hyperparameter tuning (assessed for all combinations of selected settings) and feature selection: models were trained and tested on whole expression data, iteratively reduced to the top half important genes down to <=100 genes. Models were assessed by test accuracy and area under receiver operator characteristic curve (AUC-ROC), then hyperparameter-tuned models and respective gene lists were assessed over 50 trials (random initializations and data splits). Results: Hyperparameter-tuned RF and BSVCs models with selected gene lists significantly outperformed LR in 6/6 tasks, while GB outperformed LR in 2/6 tasks. Example highest performing models (test accuracy [95% CI], AUC-ROC [95% CI]) were: For Affymetrix N, BSVCs (0.893 [0.872-0.915], 0.95 [0.931-0.969] outperformed LR (0.49 [0.453-0.527], 0.476 [0.435-0.517]). For TCGA HPV- N, BSVCs (0.832 [0.818-0.847], 0.892 [0.877-0.907]) outperformed LR (0.581 [0.568-0.595], 0.615 [0.6-0.631]). For TCGA HPV+ T, RF (0.835 [0.815-0.854], 0.954 [0.942-0.965]) outperformed LR (0.669 [0.645-0.694], 0.76 [0.732-0.789]). Conclusions: Utilizing this hyperparameter tuning and feature selection approach, we have identified machine learning techniques that can be used to analyze high dimensional data of small and large genomic expression datasets to predict clinical features and identify predictive genes for further analysis, applicable to head and neck cancer, a relatively rare malignancy (about 3% of total cancers). Citation Format: Nathan DeMichaelis, Amrit Menon, Dalton Schutte, Frank G. Ondrey. Utilizing machine learning to predict clinical staging with head and neck cancer gene expression datasets [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2400.

Abstract 7426: Leveraging deep learning to enable precision medicine via the IMPROVE benchmarking framework

Abstract The rise of targeted therapies for cancer treatment has fueled the development of algorithms that take the molecular profile (e.g. mutations, transcripts, etc.) of a tumor and predict the effect a drug has on this tumor. Recent advancements in deep learning have enabled the prediction of a cell’s response to untested, potentially novel drugs using molecular measurements [1]. To assess the clinical reliability of these algorithms, we first need to measure their performance across a range of data and model systems (e.g. cell lines, organoids, etc.). Toward this end, we introduce a benchmarking framework that enables the comparison of deep learning models for drug response across both cell line and organoid datasets. We built a fully provenanced, federated data collection pipeline that standardizes gene names, chemical structure information, and recalculates drug response curves for 10 different datasets from 4 different model system types, each accessible via our coderdata Python package (https://pnnl-compbio.github.io/coderdata/) which properly formats and processes the data for machine learning ingest. We integrated this package into the IMPROVE benchmarking framework (https://github.com/JDACS4C-IMPROVE) to evaluate the performance of eight machine learning algorithms that either leverage gene expression data alone [2-5], or with copy number [6, 7], mutation[8], and methylation data[9] added. For each algorithm we calculate the performance when using different data splitting methods as well as within-dataset and cross-dataset performance, including multiple model system types. To date, we have collected over 4, 100, 000 drug-cell datapoints across ∼55, 000 drugs and ∼1600 samples and 4 types of model system. Using our framework we evaluate (1) the effects of proper data splitting in cross-dataset analysis within a single dataset, (2) cross-dataset predictions within model systems, and (3) the ability to predict across different model systems. We are currently extending the organoid data to better enable the benchmarking of tools across the model systems and add algorithms to our benchmarking framework. We plan to use this framework to perform robust assessment of clinical applicability of AI/ML in precision medicine.

Abstract 3969: Quantitative spatiotemporal analysis of patient-derived organoid-based tumor-immune microenvironment model in lung cancer for immunotherapy response prediction

Abstract The tumor-immune microenvironment (TiME) plays a crucial role in cancer progression and treatment response, yet establishing physiologically relevant ex vivo models that accurately recapitulate TiME remains challenging. Here, we present a novel multi-modal analytical platform using lung cancer patient-derived organoids (PDOs) and autologous lymphocytes to model TiME, aiming to predict immunotherapy responses and guide treatment strategies. We developed an integrated analytical platform combining: (1) PDO-lymphocyte co-culture system using autologous cells isolated from lung cancer patients, (2) multiplexed immunofluorescence imaging for spatial analysis of immune cell distribution, (3) label-free holotomography for real-time visualization of TiME dynamics, and (4) comprehensive genomic profiling. The formation and architecture of TiME were validated through quantitative spatiotemporal immunophenotyping. Machine learning algorithms were employed to integrate multi-modal data and identify predictive signatures for treatment responses. Our integrated platform successfully reconstituted patient-specific TiME models that maintained key molecular and cellular characteristics of original tumors. Real-time holotomographic imaging revealed distinct patterns of immune cell migration and tumor-immune cell interactions. Quantitative spatiotemporal analysis demonstrated specific immune response signatures correlating with clinical outcomes. Integration of multi-modal data through machine learning algorithms showed significant predictive value for immunotherapy responses. The platform successfully identified differential drug responses in patient-specific models, correlating with clinical treatment outcomes. We established a novel multi-modal analytical platform for studying tumor-immune interactions in patient-specific models. The integration of spatial, temporal, and molecular analyses provides comprehensive insights into TiME dynamics and treatment responses. This platform shows promise as a predictive tool for immunotherapy outcomes and could serve as a personalized drug screening system, presenting new opportunities for patient stratification and therapeutic strategy optimization in lung cancer treatment. Citation Format: Mahn Jae Lee, Joo-Eun Lee, Dahye Lee, Duk Ki Kim, Da Hyun Kang, Chaeuk Chung. Quantitative spatiotemporal analysis of patient-derived organoid-based tumor-immune microenvironment model in lung cancer for immunotherapy response prediction [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3969.

Research on AI in Psychological Intervention in Adolescent Learning

In recent years, China's artificial intelligence has developed rapidly, with many policies introduced to encourage the growth of the AI industry. Moreover, it actively promotes the sustainable development of artificial intelligence. At the same time, the artificial intelligence industry has become the driving force of a new round of education revolution, actively promoting the transformation and upgrading of traditional education and giving birth to many new education industries, which is crucial to the development of educational psychology. This paper aims to show how adolescents can use artificial intelligence technology to build a personalized learning psychological intervention model to improve their learning motivation. To this end, self-determination and self-efficacy theories will be applied to answer these questions. By integrating the existing research and prospects of multiple researchers, this paper examines how adolescents can use artificial intelligence to boost their learning motivation. The findings demonstrate how teenagers can use artificial intelligence tools, like natural language processing and machine learning algorithms, to construct personalized learning psychological intervention models to improve their learning motivation, self-efficacy, and academic performance.

Predicting COPD Risk Using Machine Learning Algorithms

Chronic Obstructive Pulmonary Disease (COPD) is a progressive lung condition that affects quality of life and contributes to high global mortality. Early detection is vital but often hindered by complex, traditional diagnostic methods. This project presents a non-invasive COPD prediction system using sound analysis and machine learning. By examining respiratory sounds like coughs and wheezes from simple audio recordings, the system extracts key acoustic features to predict COPD risk. Developed as a user friendly web tool, it allows real-time predictions from uploaded recordings, making it a cost-effective and scalable solution especially in remote or low-resource areas. The model also offers interpretability, supporting informed clinical decisions and proactive care. . Key Words: Chronic Obstructive Pulmonary Disease (COPD), COPD Prediction System, Machine Learning, Healthcare Analytics, Data Classification, Predictive Modeling

Hand Gesture Robotic Arm

The paper represents the evolution of a hand gesture controlled robotic arm, designed to improve human-robot interaction by providing a natural, automotive interface. Joysticks or remote controllers which are traditional robotic control methods, require crucial user training and can be hard and complex to operate. The proposed system uses a machine vision base webcam that captures user hand movements, the hand gesture commands are processed using machine learning algorithm in Python, which generate commands to robotic arm is controlled via Arduino. The system is trained for real-time responsiveness, to understand a wide variety of hand gestures. Replication accuracy of up to 95% can achieved.

The Role of Google Earth Engine in Flood Disaster Management: A Review of Capabilities and Challenges

This article provides a comprehensive overview of flood mapping using Google Earth Engine (GEE), emphasizing its significance in disaster management and mitigation. Flood mapping is crucial for understanding risks and preparing for flood events, which are becoming more frequent and severe due to climate change. GEE's advanced capabilities in processing and analyzing vast datasets of satellite imagery offer a potent tool for rapid and accurate flood mapping. Studies reviewed demonstrate GEE's application in various contexts, highlighting its effectiveness in generating timely and precise flood maps using methodologies like image processing techniques, machine learning algorithms, and hydrological modeling. Challenges such as technical complexity and data availability are noted, alongside the potential for future advancements integrating AI and big data analytics to enhance flood risk management strategies. The paper underscores GEE's role in advancing flood mapping efforts, contributing to better preparedness and response to flood disasters worldwide.

Abstract 3317: IFITM3: A predictive biomarker for therapeutic outcomes in advanced hepatocellular carcinoma

Abstract Background: Hepatocellular carcinoma (HCC) is a major global health challenge, with China bearing a significant burden due to its high incidence and mortality rates. Vascular invasion (VI), particularly microvascular invasion (MVI), is a critical prognostic factor in HCC , profoundly influencing treatment decisions. Despite its clinical importance, reliable and non-invasive biomarkers for stratified prediction and therapeutic response remain scarce. Extracellular vesicles (EVs) are emerging as promising tools in cancer progression and therapy assessment. Methods: Sixty-five patients with China Liver Cancer (CNLC) stage III HCC treated with hepatic artery infusion chemotherapy (HAIC) plus sintilimab and a bevacizumab biosimilar (IBI305) were stratified based on RECIST criteria and surgical status for survival analysis and Kaplan-Meier (KM) curve construction. Univariate and multivariate regression analyses were employed to identify independent predictors of surgical eligibility. Cellular and molecular investigations followed to uncover potential biomarkers. scRNA-seq data (GEO: GSE242889), bulk RNA-seq data (GEO: GSE19977, TCGA), and hdWGCNA analysis were utilized to identify hub genes linked to MVI. TCGA data were divided into training and validation sets at a 7:3 ratio. Machine learning algorithms (LASSO regression, SVM-RFE, random forest) identified high-risk MVI genes. A prognostic nomogram model was constructed and significant predictors were explored with its performance assessed via ROC curves. Immunohistochemical staining and exosomal IFITM3 analysis were applied to evaluate gene expression across treatment stages. Results: Data from 65 patients (April 2021 to July 2023) with a median age of 57 years, predominantly male (83.1%), were analyzed. Patients who underwent surgery (30.8%) or achieved partial response (PR) per RECIST (38.5%) demonstrated significantly improved overall survival. Multivariate regression indicated that patients with VI, distant metastasis, ascites, and Child-Pugh grade B as factors reducing surgical eligibility, with VI as an independent predictor. Six MVI high-risk genes (VTN, DYNC1I2, HABP2, IFT43, IFITM3, NDUFS4) were identified, among which IFITM3 emerged as a key prognostic marker. High IFITM3 expression was associated with shorter survival (P < 0.001). The nomogram model showed strong predictive accuracy for survival (AUC1=0.641, AUC2=0.613, AUC2.5=0.579). Immunohistochemical analysis and exosomal analysis further validated IFITM3's role in influencing therapeutic outcomes. Conclusions: In summary, IFITM3 levels are closely linked to therapeutic outcomes in HCC patients. Non-invasive blood-based detection of IFITM3 holds promise for enhancing therapeutic stratification. Further research may refine prediction models and deepen understanding of IFITM3's role in poor therapeutic outcomes in cancer. Citation Format: Yinchao Zhang, Linze Xu, Jin Zhang, Dandan Wu, Huikai Li, Yueguo Li, Yang Liu. IFITM3: A predictive biomarker for therapeutic outcomes in advanced hepatocellular carcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3317.

Abstract 3714: Microscopic mayhem: Serial killing and other immune cell dynamics captured in situ

Critical to advancing immunotherapy and cell therapy in cancer is developing a deeper understanding pf the dynamics of immune cell-mediated cytotoxicity. The results from the multidisciplinary effort reported here include numerous movies of immune cell-mediated cytotoxicity with striking examples of serial killing, foraging, path-tracking, cytokine gradients at tumor margins, and killing dynamics, in some cases revealing peak apoptotic signatures just minutes after T Cell engagement. In vitro studies of immune cell killing are traditionally performed using time-lapse imaging and biochemical assays, but these methods are often limited by spatial and temporal resolution, throughput, and the ability to extract the dynamics of cellular interactions. This study integrates high-resolution and high-speed laser scanning confocal microscopy with artificial intelligence (AI), and machine learning (ML) approaches to provide a high-resolution data-driven analysis of immune cell killing dynamics in vitro. We have engineered a perfusion-enabled 3D culture system integrated microscopy to assess cellular dynamics for extended periods of time. Perfusion culture maintains the interstitial flow of liquid culture media, clearing the microenvironment of toxic metabolites and reactive oxygen species. This platform uses a Liquid-Like Solids (LLS) to mimic the transport dynamics of a capillary bed. Integrated microscopy allows in situ quantification of spatiotemporal cytokine concentrations, immune cell tracking, immune cell killing dynamics, and invasion dynamics. In one example, we used Human CAR T cells with an anti-cd19.28z CAR that included a fluorescent mCherry protein fused to the carboxy terminus of the CAR vs. target cancer cells (Burkitt lymphoma) that express GFP. The media contained a 0.1% concentration of Propidium Iodide as a terminal maker of cell death. Such multi-channel data highlights the critical stages of immune-mediated cytotoxicity; however, the complexity and volume of confocal imaging data necessitates advanced computational tools for analysis. The ML algorithm used in this study comes from computer vision using object detection, which is a cornerstone technique in medical imaging. We utilized the YOLOv8X model, which is distinguished by its high accuracy and efficiency in object detection tasks. The YOLO (You Only Look Once) architecture is designed to predict both bounding boxes and class probabilities in a single forward pass, making it exceptionally fast and suitable for real-time applications. Finally, by integrating robotics, imaging, and AI/ML detection we are able to capture immune cells of interest in situ. This integrated analysis of complex imaging data from confocal microscopy has been significantly accelerated, enabling the identification of subtle patterns and dynamic interactions between immune cells and their targets, and the actual collection of cells of interest. Citation Format: W. Gregory Sawyer, Diego Pedro, Alfonso Pepe, Reginald Atkins, Chandler Gatenbee, Luis Felipe, Carlos Garcia Fernandez, Soraya Zorro Manrique, Said Cifuentes Maury, Athena Tsingelis, Duy Nguyen, Alexander Anderson, Erin George, Gilmer Valdes, Fredrick Locke, Patrick Hwu. Microscopic mayhem: Serial killing and other immune cell dynamics captured in situ [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3714.

The Impact of Machine Learning on Future Defense Strategies

Machine learning (ML) is increasingly pivotal in shaping future defense strategies, offering advancements across critical domains such as cybersecurity, resource allocation, and autonomous systems. This research explores the multifaceted impact of ML in defense, focusing on three primary areas: enhancing threat detection and mitigation in cybersecurity, optimizing resource allocation and logistics in military operations, and navigating the ethical and strategic implications of autonomous defense systems. The study utilizes qualitative research methods, particularly through secondary data analysis from government reports, academic publications, industry white papers, and ethical frameworks. Findings indicate that ML algorithms significantly bolster threat detection by leveraging Anomaly Detection Theory and Game Theory, thereby enhancing responsiveness to cyber threats. In military logistics, ML models informed by Operations Research and Supply Chain Management Theory optimize resource distribution, improve operational efficiencies, and support mission readiness. Ethically, the deployment of ML in autonomous defense systems prompts considerations of moral responsibilities, biases, and strategic risks, necessitating comprehensive governance frameworks. In conclusion, while ML offers transformative potential in defense strategies, effective implementation requires robust ethical guidelines, strategic foresight, and interdisciplinary collaboration to mitigate risks and maximize benefits.

Abstract 6210: The predictive performance of risk prediction models in determining lung cancer incidence in Western and Asian countries: A systematic review and meta-analysis

Abstract Introduction: Numerous prediction models have been developed to identify high-risk individuals for lung cancer screening, aimed at improving early detection and survival rates. However, there has been no comprehensive review and meta-analysis of model performance across different sociocultural contexts. Therefore, this review aims to systematically examine the performance of lung cancer risk prediction models in Western and Asian populations. Methods: PubMed and EMBASE were searched from inception to January 2023. Studies published in English and proposed a validated model on human populations with well-defined predictive performances were included. Two reviewers independently screened titles and abstracts and the Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to appraise study quality. A random-effects meta-analysis was performed and a 95% confidence interval (CI) for model performance was provided. Between-study heterogeneity was adjusted for using the Hartung-Knapp-Sidik-Honkman test. Results: Fifty-four studies were included, with 42 from Western and 12 from Asian countries. Most Western studies were designed for ever-smokers (19/42; 45.2%) and the general population (17/42; 40.5%), and only two Asian studies developed models exclusively for never-smokers. Across both Western and Asian prediction models, the three most consistently included risk factors are age, sex, and family cancer history. In 45.2% (19/42) of Western and 50.0% (6/12) of Asian studies, models incorporated both traditional risk factors and biomarkers; 14.8% (8/54) directly compared their biomarker-based models with those incorporating only traditional risk factors, demonstrating better discrimination. Machine-learning algorithms were applied in eight Western models and two Asian models. External validation of PLCOM2012 (AUC=0.748; 95% CI: 0.719-0.777) outperformed prediction models such as Bach (AUC=0.710; 95% CI: 0.674-0.745) and Spitz models (AUC=0.698; 95% CI: 0.640-0.755). Conclusion: Despite showing promising results, the majority of Asian risk models in our study lack external validation. Our review also reveals a significant gap in prediction models for never-smokers. Future research should externally validate existing Asian models or incorporate relevant Asian risk factors in widely-used Western models (PLCOM2012) to account for their unique risk profiles and lung cancer progression patterns. Citation Format: Yah Ru Juang, Lina Ang, Wei Jie Seow. The predictive performance of risk prediction models in determining lung cancer incidence in Western and Asian countries: A systematic review and meta-analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6210.

Abstract 3338: A CpG methylation based epigenetic signature for recurrence prediction in stage II/III colorectal cancer: The CENSURE Study

Introduction: Adjuvant chemotherapy (ACT) is the standard of care for patients with stage III or high-risk stage II colorectal cancer (CRC) following curative-intent resection. However, over 30% of these patients experience recurrence despite treatment, highlighting the urgent need for developing predictive biomarkers of response to ACT. This study sought to identify epigenetic biomarkers for predicting recurrence in stage II/III CRC. Patients and Methods: This study included three phases: discovery, training, independent testing. In the discovery phase, we performed genomewide methylation profiling analysis between recurrent (recurrence-free survival, RFS < 5 years) and non-recurrent patients with CRC. A LASSO-based machine learning algorithm was employed to refine and optimize the candidate CpG sites. Subsequently, we developed and tested a methylation biomarker panel by quantitative pyrosequencing assays, followed by eXtreme Gradient Boosting (XGB) machine-learning algorithm in both the training (n = 236) and testing (n = 175) cohorts. Finally, we established the CENSURE (CpG mEthylatioN-based epigenetic SignatURE) model, which combined the CpG methylation panel and carcinoembryonic antigen (CEA) levels, to predict recurrence in stage II/III CRC. Results: A panel of 8 CpG candidates identified in the discovery phase was refined to 6 using XGB in the training cohort. The 6-CpG panel demonstrated robust ability to predict recurrence, with AUC values of 91.2% and 79.3% in the training and testing cohorts, respectively. Multivariate Cox regression analysis revealed that CEA ≥ 5 ng/mL (HR 2.92; 95% CI: 1.81-4.70) and the CpG panel (HR 19.5; 95% CI: 8.78-43.1) were independent predictors of recurrence. Building on this, we developed the CENSURE model, which integrates the CpG panel with CEA. The CENSURE model achieved AUC values of 95.2% and 86.1%, with sensitivities of 83.9% and 83.3% in the training and testing cohorts, respectively. CENSURE high-risk patients exhibited significantly higher 5-year (71.5% vs 8.0%) and 2-year recurrent rates (48.0% vs 5.4%). Notably, the CENSURE model outperformed CEA risk stratification alone, correctly predicting an additional 46.5% (40 out of 86) of recurrence cases in the entire study cohort. Conclusions: In conclusion, the CENSURE assay represents a novel potential advance with a significant clinical impact for postoperative recurrence prediction in stage II/III CRC. Citation Format: Caiming Xu, Silei Xu, Joan Maurel, Man Li, Ajay Goel. A CpG methylation based epigenetic signature for recurrence prediction in stage II/III colorectal cancer: The CENSURE Study [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3338.

Genomics and the early diagnosis of lung cancer.

Lung cancer (LC) remains the leading cause of cancer-related mortality worldwide, with most cases diagnosed at advanced stages, resulting in poor survival rates. Early detection significantly improves outcomes, yet current screening methods, such as low-dose computed tomography (LDCT), are limited by high false-positive rates, radiation exposure, and restricted eligibility criteria. This review highlights the transformative potential of genomic and molecular technologies in advancing the early detection of LC. Key innovations include liquid biopsy tools, such as circulating tumor DNA (ctDNA) and cell-free DNA (cfDNA) analysis, which offer minimally invasive approaches to detect tumor-specific genetic and epigenetic alterations. Emerging biomarkers, including methylation signatures, cfDNA fragmentomics, and multi-omics profiles, demonstrate improved sensitivity and specificity in identifying early-stage tumors. Advanced platforms like next-generation sequencing (NGS) and machine-learning algorithms further enhance diagnostic accuracy. Integrated approaches that combine genomic data with LDCT imaging and artificial intelligence (AI) show promise in addressing current limitations by improving risk stratification and nodule characterization. The review also explores multi-cancer early detection assays and precision diagnostic strategies tailored for diverse at-risk populations. By leveraging these advancements, clinicians can achieve earlier diagnoses, reduce unnecessary procedures, and ultimately decrease LC mortality.