Implementation of the Random Forest Algorithm with Optuna Optimization in Lung Cancer Classification

Lung cancer, especially non-small cell lung cancer (NSCLC), is one of the most-deadly malignancies worldwide. Lung cancer has a worse 5-year survival rate than many primary malignancies. Thus, the early detection and prognosis prediction of lung cancer are crucial. The early detection and prognosis prediction of lung cancer have improved with the widespread use of artificial intelligence (AI) technologies. This meta-analysis examined the accuracy and efficacy of AI-based models in predicting lymph node metastasis (LNM) in NSCLC patients using imaging data. Our findings could help clinicians predict patient prognosis and select alternative therapies. We searched the PubMed, Web of Science, Cochrane Library, and Embase databases for relevant articles published up to January 31, 2024. Two reviewers individually evaluated all the retrieved articles to assess their eligibility for inclusion in the meta-analysis. The systematic assessment and meta-analysis comprised articles that satisfied the inclusion criteria (e.g., randomized or non-randomized trials, and observational studies) and exclusion criteria (e.g., articles not published in English), and provided data for the quantitative synthesis. The quality of the included articles was assessed using the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2). The pooled sensitivity, specificity, and area under the curve (AUC) were used to evaluate the ability of AI-based imaging models to predict LNM in NSCLC patients. Sources of heterogeneity were investigated using meta-regression. Covariates, including country, sample size, imaging modality, model validation technique, and model algorithm, were examined in the subgroup analysis. The final meta-analysis comprised 11 retrospective studies of 6,088 NSCLC patients, of whom 1,483 had LNM. The pooled sensitivity, specificity, and AUC of the AI-based imaging model for predicting LNM in NSCLC patients were 0.87 [95% confidence interval (CI): 0.80-0.91], 0.85 (95% CI: 0.78-0.89), and 0.92 (95% CI: 0.90-0.94). Based on the QUADAS-2 results, a risk of bias was detected in the patient selection and diagnostic tests of the included articles. However, the quality of the included articles was generally acceptable. The pooled sensitivity and specificity were heterogeneous (I2>75%). The meta-regression and subgroup analyses showed that imaging modality [computed tomography (CT) or positron emission tomography (PET)/CT], and the neural network method model design significantly affected heterogeneity of this study. Models employing sample size data from a single center and the least absolute shrinkage and selection operator (LASSO) method had greater sensitivity than other techniques. Using the Deek' s funnel plot, no publishing bias was found. The results of the sensitivity analysis showed that deleting each article one by one did not change the findings. Imaging data models based on AI algorithms have good diagnostic accuracy in predicting LNM in patients with NSCLC and could be applied in clinical settings.

The Center for Global Early Detection at MD Anderson has initiated two longitudinal early detection studies: LEAP (Lung Cancer Early Detection and Prevention, 2013-2019) for lung cancer screening and MERIT (Mammography Early Detection Risk Assessment and Imaging Technologies, 2017-ongoing) for breast cancer screening. Both cohorts enrolled subjects based on US standard-of-care guidelines at the time of enrollment for low-dose CT screening and mammography, respectively. The cohorts collected detailed health data from questionnaires, imaging data, and blood specimens at the time of screening. In addition to information on the screened-detected cancers, follow-up information was collected on any other types of cancers that are diagnosed after enrollment. Here, we evaluate the incidence of other cancers without available screening to assess the potential impact of multi-cancer early detection studies for these two types of screening populations. The two studies enrolled approximately 9,300 subjects (N = 2853 for LEAP; N = 6,400 for MERIT) with median follow up times of 2.99 years (IQR: 1.09-4.08) for LEAP and 2.05 years (IQR: 0.09-3.01) for MERIT. Currently, there are more than 800,000 aliquots of banked biospecimens from both cohorts. The incidence of lung cancer for the LEAP cohort undergoing low-dose CT screening was 0.91% per year. Cancers other than lung cancer were also diagnosed after enrollment including prostate, breast, bladder, head, and neck, colorectal, lymphoma, endocrine, gynecologic, and kidney cancers. Overall, the incidence of other cancers in LEAP was high for those cancers without available screening - 0.49% per year excluding breast, prostate, and colorectal cancers. The incidence of breast cancer for women in LEAP was lower than expected, 0.17% per year, likely a result of underscreening by mammography. For the MERIT cohort, the incidence of breast cancer was 0.89% per year. Cancers other than breast were diagnosed in MERIT after enrollment including gynecologic, melanoma, lung, colorectal, endocrine, bladder, and gastrointestinal cancers. The incidence of other cancers in MERIT was 0.55% per year, excluding colorectal cancer. We conclude that the incidence of cancers without a current screening modality in LEAP and MERIT is sufficiently high (0.49-0.55% per year) to justify the application of multi-cancer screening. Even for cancers with standard-of-care screening, a multi-cancer blood-based biomarker test could be useful given the possibility of underscreening, as we observed for breast cancer in the LEAP cohort. Citation Format: Jennifer Dennison, Ehsan Irajizad, Edwin Ostrin, Myrna Godoy, Olena Weaver, Jessica Leung, Samir Hanash. Application of multi-cancer early detection to lung and breast cancer screening populations. [abstract]. In: Proceedings of the AACR Special Conference: Precision Prevention, Early Detection, and Interception of Cancer; 2022 Nov 17-19; Austin, TX. Philadelphia (PA): AACR; Can Prev Res 2023;16(1 Suppl): Abstract nr P057.

Lung cancer remains one of the deadliest cancers worldwide, with survival rates significantly dependent on early detection. Traditional diagnostic methods, while effective, often identify lung cancer in advanced stages, limiting treatment options. Recently, artificial intelligence (AI) and deep learning (DL) models have gained attention for their potential to assist in early lung cancer diagnosis through radiological data, particularly computed tomography (CT) and X-ray imaging. This review examines recent advancements in AI and DL models specifically applied to early lung cancer detection, offering a comprehensive look at model architecture, data preprocessing techniques, and performance metrics. AI models, particularly those using convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have shown significant promise in improving diagnostic accuracy. In studies published over the past two years, CNN-based models have achieved sensitivity rates upwards of 90% when applied to large datasets, such as the LIDC-IDRI, a publicly available database containing thousands of annotated lung scans. Furthermore, hybrid approaches combining AI models with radiologist expertise have demonstrated reduced false positives, a frequent challenge in automated image analysis. A critical factor in these advancements has been data quality and volume. Large, annotated datasets with high-resolution images have enabled more robust training and validation, essential for refining these models. However, challenges remain, particularly regarding the standardization of data across institutions, variations in scanner quality, and ethical concerns surrounding patient privacy. The review also highlights studies exploring multi-modal approaches, integrating radiological data with clinical records and genetic markers to create more personalized diagnostic tools. These multi-modal methods have the potential to improve predictive accuracy further, though they currently require more extensive validation. Overall, while AI and DL technologies offer transformative potential for early lung cancer detection, widespread implementation depends on continued research in model accuracy, data standardization, and ethical safeguards. This paper concludes by emphasizing the need for collaboration across disciplines—AI researchers, radiologists, and oncologists—to refine these models into reliable tools that can be integrated seamlessly into clinical workflows for timely and accurate lung cancer diagnosis.

8553 Background: Lung cancer early detection improves survival, but risk-based low-dose CT screening (LDCT) only identifies a minority of patients. We implemented an ILNP in a community healthcare system, and evaluated its risks and benefits. Methods: Patients with lung lesions on routinely-performed radiologic studies were flagged by radiologists and triaged using evidence-based guidelines. We tracked demographics, clinical characteristics, procedures, complications, and health outcomes. We analyzed ILNP subjects’ eligibility for LDCT by National Lung Screening Trial (NLST), Center for Medicaid Services (CMS), NEderlands Leuvens Screening ONderzoek (NELSON), National Comprehensive Cancer Network (NCCN) Risk Groups 1 and 2 (screening recommended), NCCN Risk Group 3 (screening not currently recommended), and US Preventive Services Task Force (USPSTF) criteria from 2013 and 2020. Statistical analysis used the chi-square test and Kaplan Meier method. Results: From 2015-2020, 13,710 patients were evaluated in the ILNP program: median age, 64 years; 42% male; 65% White, 29% Black; 667 (4.9%) were diagnosed with lung cancer. Lung cancers diagnosed from ILNP were 39% adenocarcinoma / 20% Squamous Cell with clinical stage distribution 49% I, 8% II, 17% III, and 16% IV. 832 (6.1%) had invasive diagnostic testing- CT-guided biopsy (50%), bronchoscopy (30%), and/or EBUS (26%); 11% of the 832 had >1 invasive diagnostic test. The most common complications from invasive testing were pneumothorax and chest tube placement. Only 11%-20% of all ILNP patients would have been eligible for LDCT. In ILNP patients diagnosed with lung cancer, only 33% were eligible for screening by NLST criteria; the proportion increased substantially when USPSTF 2020 or NCCN Group 2 criteria were applied (Table). Compared to NLST, NCCN Group 2 criteria increased screening eligibility among cancer patients by 22% (from 33% to 55%), while only increasing screening eligibility by 6% (from 8% to 14%) in non-cancer patients. Aggregate 1-year and 3-year survival rates for lung cancer patient diagnosed through ILNP were 76% (95% CI: 73, 80) and 64% (95% CI: 59, 69). Conclusions: The ILNP identified early-stage lung cancer more frequently than most LDCT programs, with promising survival rates. The majority of subjects with lung cancer were not eligible for LDCT, we still need to optimize risk-based screening criteria. Even with new, expanded criteria for LDCT, structured ILNP is necessary to expand early detection of lung cancer.[Table: see text]

The examination and treatment of lung cancer in its early stages rely heavily on diagnostic imaging technologies. Various imaging procedures used in healthcare, such as computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET), chest x-ray, and molecular imaging (MI) approaches, have been extensively studied for the detection and classification of lung cancer. However, some of these systems have limitations, such as their inability to accurately classify lung cancer images, rendering them unsuitable for use in individuals with other illnesses. There is a critical need for the development of a robust and precise method for the early detection of lung cancer. One of the fastest-growing areas in health imaging is deep learning (DL), which involves rapidly developing techniques that utilize texture and clinical-image-based information. DL-based clinical imaging technologies may assist physicians in recognizing and classifying lung cancers more easily and accurately. In this paper, a deep ensemble model is proposed for detecting and classifying lung cancer based on three DL classifiers: MobileNet, ResNet50, and XceptionNet. These DL models are trained on a publicly available lung cancer dataset containing normal, benign, and malignant images. The performance of these baseline classifiers can be measured in terms of accuracy, precision, and recall. The proposed ensemble model accurately detects and classifies lung cancer images. The final results show classification accuracy scores of 28.00%, 13.00%, and 36.00% for the baseline classifiers MobileNet, ResNet50, and XceptionNet, respectively, over the dataset. The proposed ensemble model outperforms the baseline classifiers, achieving a classification accuracy score of 99.00%. Finally, the performance of the proposed ensemble model is compared with those of existing state-of-the-art methods.

126 Background: Lung cancer (LC) causes the most cancer deaths in the US, but early detection improves outcomes. Uptake of Low-Dose CT screening (LDCT) has been low, contributing to race-based disparities. Programmatic management of incidentally detected pulmonary nodules (PN) provides a complementary avenue for early detection which could address disparities. In our PN cohort, Detecting Early LC in the Mississippi Delta (DELUGE), we provide broad access to early detection across a region fraught with geographic, socioeconomic, and race-based disparities. However, many detected PN will not result in LC. After successful implementation at the program level, we seek to evaluate individual-level factors important for early-detection. We evaluated how individuals with small PN identified in DELUGE differ between those who were and were not diagnosed with LC. Methods: DELUGE is a prospective cohort study where we systematically manage incidentally detected PN with trained navigators and evidence-based risk stratification. We constructed DELUGE in a community-based healthcare system across the mid-south with some of the highest per-capita LC incidence and mortality rates in the US. We evaluated all PN ≤30mm in DELUGE from 2015-2022. We compared clinical and demographic characteristics between individuals who were or were not diagnosed with LC within 24 months. Primary analysis used logistic regression, reporting odds ratios (OR) with 95% confidence intervals. Results: From 2015-2022, 17,945 individuals with PN ≤30 mm (71% of the total cohort of 25,339) were identified and enrolled in DELUGE. They had median age of 65 years, were 56% female, and 27% Black / 69% White. 853 (4.8%) patients were diagnosed with LC within two years. PN size significantly predicted LC diagnosis. For every 1 mm increase in nodule size the odds of LC increased by 15% (OR: 1.15 [1.14-1.16]); PN in an upper lobe (OR: 2.0 [1.7-2.2]) and those with cavitation also increased the odds of LC (OR: 2.9[1.9-4.2]). Sex, rural/urban residence, and number of PN were not associated with LC. The most predictive individual level characteristics were age, prior cancer history (OR: 1.3 [1.1-1.5], family cancer history (OR: 1.7 [1.5-2.0], COPD (OR: 3.3 [2.9-3.8]), and smoking history (OR: 6.6 [5.3-8.4]). In the multiple variable model, the most significant associations with LC diagnosis included PN size, age, insurance, family cancer history, COPD, and smoking history (all p<0.0001). This model could provide estimated sensitivity 0.81 and specificity 0.81 for predicting LC (AUC= 0.88). Conclusions: Factors measurable at baseline have predictive ability, and predictive modeling may be useful in management of small PN. More focused follow-up could improve programmatic efficiency, targeting the alleviation of geographic, socioeconomic, and race-based health disparities in this high poverty region of the US.

Lung cancer is a highly lethal global problem. The disease is too often detected only at a terminal stage, when treatment by even the best current methods is insufficient. Better methods of early lung cancer detection are fundamental for a shift to the routine management of truly early (preinva-sive) lung cancer. This should be based on two main concepts: 1) A large body of evidence is accumulating which suggests that invasive lung cancer is the end result of a multistep process, in which progressive molecular changes accompany or even precede morphological changes. Genetic changes, because of their exquisite specificity or selectivity, are ideally suited as markers of premalignancy. However, a knowledge of the genetic etiology is required before the markers can be used. It is usually assumed that a phenotypically normal cell is completely genetically normal. 2) The entire epithelium is the target of carcinogens and this is the basis of the field effect concept. Early detection should therefore aim to identify the field effect and devise suitable methods of intervention. Our knowledge is still rudimentary: the clonal/multifocal relationship between pre- and postinvasion is still being investigated. Could a random biopsy yield information for the rest of the field? This is still a matter of discussion. Multiple biopsies should be performed in order to evaluate how broad the cancerization field is as regards genetic damage.

Introduction: Early detection of lung cancer by screening of high risk populations (identified by epidemiological and life-style factors) has the potential to save many lives. However, effective screening is reliant on minimally invasive techniques, such as CT screening, bronchioalveolar lavage (BAL) and blood tests, and the identification of suitable biomarkers. CT screening is effective in reducing mortality, but generates a large proportion of indeterminate nodules that must be further characterised. MicroRNAs (miRNA) have great potential as biomarkers due to their tissue-specific and cancer-specific expression patterns. We have identified tumour-specific miRNAs for non-small cell lung cancer (NSCLC), using a combination of screening on TaqMan microRNA TLDA cards and validation with qRTPCR assays, with the aim of utilising these as biomarkers in the early detection setting. Methods: Our sample group consisted of 31 frozen samples from 20 Liverpool Lung Project (LLP) NSCLC patients, including 10 adenocarcinomas (Ad), 10 squamous cell carcinomas (SCC) & matched normal tissue. Two further validation sets consisted of equal numbers of Ad and SCC tumour/normal pairs (124 in total). MiRNA was prepared from tumour and normal specimens using Qiagen MicroRNeasy kits. Reverse transcription and pre-amplification was performed using Applied Biosystems MegaPlex Pools and miRNAs were quantified on a 7900HT Real-Time PCR System with TaqMan Array Human MiRNA Card Set v3.0 (covering 754 human miRNAs). Ct values were exported using SDS v2.3 data and RQ Manager software and further analysed in Bioconductor. Validation qRTPCR was performed with individual miRNA assays, following reverse transcription with MegaPlex pools. Results: When Benjamin-Hoechst-adjusted-p value <0.05 was used as a cut-off, of the 754 miRNA targets, 68 miRNAs were upregulated and 8 were downregulated with >4.0 fold-change in the cancer group. A subset of 22 miRNAs including miR-34a, miR-96, let-7g and miR-183 was identified with the greatest expression in tumours. Differential expression of all 22 miRNAs was confirmed in an independent set of 24 tumour/normal pairs. Using these 22 validated miRNAs we performed discriminative modelling and identified a model based on just 8 markers that gave a specificity of 100% and a sensitivity of 98%. This panel was validated, with 97% specificity and 91% sensitivity, in a 2nd independent sample set containing 48 tumours and paired normal samples. Conclusion: A number of miRNAs was identified that showed good discriminatory power individually, but greatest sensitivity and specificity when combined as an 8 member panel. The lung cancer specific miRNAs we have identified provide a potential source of early detection biomarkers. Their applicability to minimally-invasive samples is being evaluated in a range of samples including plasma and bronchial lavage. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 4133. doi:1538-7445.AM2012-4133

Lung cancer remains the leading cause of cancer-related mortality worldwide, highlighting the urgent need for earlier detection within real-world screening and patient management pathways. Recent advances in multi-omics technologies have created new opportunities for identifying biomarkers associated with early-stage lung cancer, particularly in high-risk populations under clinical surveillance. This review systematically evaluates early diagnostic biomarkers across multiple omics layers, including genomics, epigenomics, transcriptomics, proteomics, metabolomics and microbiomics. It also summarises the application of artificial intelligence (AI), particularly machine learning and deep learning approaches, for integrating and analysing complex multi-omics datasets to support biomarker discovery and clinical decision-making. Multi-omics strategies are accelerating the identification of molecular signatures relevant to early lung cancer detection. AI-driven methods enable the extraction of latent patterns from high-dimensional data, facilitating risk stratification, diagnostic refinement, histological subtyping and treatment planning. The review highlights the clinical utility of these biomarkers and their potential incorporation into screening algorithms, as well as the development of AI-based clinical decision support systems (CDSS) aligned with real-world clinical workflows. However, major barriers to clinical translation remain, including multi-centre data heterogeneity, limited model interpretability affecting clinical trust, regulatory and cost-effectiveness challenges and insufficient validation in prospective cohorts. Emerging technologies, such as single-cell and spatial multi-omics, along with federated learning frameworks, offer promising solutions to bridge the gap between computational discovery and clinical implementation. The integration of AI and multi-omics approaches has the potential to advance risk-adapted and personalised early detection strategies for lung cancer.

Accurate classification and early detection of lung cancer are crucial for effective treatment and improved patient outcomes. Although recent advances in medical imaging have improved diagnostic workflows, many traditional approaches remain limited in their ability to efficiently analyze large volumes of imaging data. Accordingly, this study proposes a deep learning framework for lung cancer classification using Digital Imaging and Communications in Medicine (DICOM) images. The proposed approach integrates an Intelligent Squirrel Search (ISS)–tuned VGG16 network (ISS-VGG16) to improve classification performance. A set of lung cancer DICOM images was obtained from an open-source dataset. Preprocessing steps, including image resizing and contrast enhancement, were applied to standardize the inputs for model training. ISS was used to optimize key hyperparameters of the VGG16 model, thereby improving classification performance. The model was implemented in Python. Experimental results indicate that the proposed ISS-VGG16 approach outperforms baseline methods, achieving a recall of 0.92, precision of 0.96, F1-score of 0.95, and overall accuracy of 0.97 on the evaluated lung cancer image dataset. These results demonstrate reliable classification performance across the defined lung cancer classes. Overall, the proposed framework provides an effective and dependable approach for lung cancer image classification.

Early Detection of Lung Cancer with Meso Tetra (4-Carboxyphenyl) Porphyrin-Labeled Sputum

Exosomal lipids for classifying early and late stage non-small cell lung cancer

Lung cancer remains the leading cause of cancer-related mortality worldwide, with poor survival rates due to late-stage detection. Current low-dose computed tomography screening faces barriers including high costs and false-positive rates reaching 24%, while artificial intelligence offers opportunities to enhance early detection through longitudinal clinical data analysis. This study developed a Multi-Channel Convolutional Neural Network (MCNN) for lung cancer risk prediction using Taiwan's National Health Insurance Research Database, encompassing 523,539 patients (2,809 lung cancer, 23,783 other cancer, and 496,947 non-cancer). The MCNN was designed as a lightweight model processing nine channels of diagnostic codes, medications, and medical orders over a three-year observation period. Systematic feature selection reduced estimated feature storage requirements by 99.8%, from approximately 1,184 GB for the full ICD feature space to approximately 2.11 GB for the selected features, while retaining clinical relevance. Model performance was assessed using stratified 10-fold cross-validation against seven machine learning baselines, and interpretability was examined through SHAP analysis. The MCNN achieved an F₁-score of 66.91%, precision of 84.47%, and recall of 59.79%. Ablation studies confirmed multi-modal integration benefits, with diagnostic codes providing primary predictive power. SHAP analysis revealed distinct temporal patterns validating the model's ability to identify pre-diagnostic phases through healthcare engagement patterns. Findings are based on internal validation within a single national database, and key risk factors such as smoking history are not captured in administrative claims data; future evaluation in independent external cohorts is therefore warranted to confirm these findings. The model's high precision minimizes false-positive rates while its computational efficiency and clinical interpretability support practical implementation as a complementary claims-based screening support tool for early cancer detection.

Background: Improving prediction models to timely detect lung cancer is paramount. Our aim is to develop and validate prediction models for early detection of lung cancer in primary care, based on free-text consultation notes, that exploit the order and context among words and sentences. Methods: Data of all patients enlisted in 49 general practices between 2002 and 2021 were assessed, and we included those older than 30 years with at least one free-text note. We developed two models using a hierarchical architecture that relies on attention and bidirectional long short-term memory networks. One model used only text, while the other combined text with clinical variables. The models were trained on data excluding the five months leading up to the diagnosis, using target replication and a tuning set, and were tested on a separate dataset for discrimination, PPV, and calibration. Results: A total of 250,021 patients were enlisted, with 1507 having a lung cancer diagnosis. Included in the analysis were 183,012 patients, of which 712 had the diagnosis. From the two models, the combined model showed slightly better performance, achieving an AUROC on the test set of 0.91, an AUPRC of 0.05, and a PPV of 0.034 (0.024, 0.043), and showed good calibration. To early detect one cancer patient, 29 high-risk patients would require additional diagnostic testing. Conclusions: Our models showed excellent discrimination by leveraging the word and sentence structure. Including clinical variables in addition to text slightly improved performance. The number needed to treat holds promise for clinical practice. Investigating external validation and model suitability in clinical practice is warranted.

Medical healthcare systems store a large amount of clinical data about patients related to their biographies and disease information. Doctors use clinical data for the early detection of diseases that helps with proper patients’ treatments to save their lives. These clinical systems are helpful in detecting cancer diseases at early stages to save people's lives. Lung cancer is the third largely spreading disease in human beings all over the globe, which may lead so many people to death because of inaccurate detection of their disease at the initial stages. Therefore, this study will help doctors and radiologists in the detection of lung cancerous and non-cancerous patients at early stages with a random forest algorithm to save patients’ lives. In this research work, a new and novel model based on random forest algorithm was employed to detect lung cancer from the Wisconsin data set. Lung cancer was detected at early stages, and it was decided whether targeted patient was cancerous or non-cancerous. This experimental outcome showed that the proposed methodology achieved an accuracy rate that was batter compared to previous studies for early detection of lung cancer.