Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 840 https://internationalpubls.com Framework for Early Detection of Lung Malignancies using Deep Learning Techniques Saba Fatima1 Mrs. MD Asma2 Dr. T.K. Shaik Shavali3 1 Research Scholar, Dept. of Computer Science and Engineering, Lords Institute of Engineering and Technology, Hyderabad, Telangana 2Associate professor, Dept. of Computer Science and Engineering, Lords Institute of Engineering and Technology, Hyderabad, Telangana 3Professor, Dept. of Computer Science and Engineering, Lords Institute of Engineering and Technology, Hyderabad, Telangana Article History: Received: 12-01-2025 Revised: 15-02-2025 Accepted: 01-03-2025 Abstract Lung cancer is a disease that has a major impact on public health. This study suggests folding networks (CNNS) and Densen's approaches to support lung cancer recognition and classification. In various areas of pattern recognition and medical imaging, CNN and Densenet have demonstrated their effectiveness. In this study, many medical lung images were created using x-rays from people with lung cancer. The results show that it can be maliciously divided using CNN and Densenet architectures with 99 parameter accuracy.8%. This study contributes to the creation of a deep learning-based system for detecting and classifying lung cancer. The results could be the basis for creating a more accurate and productive diagnostic system for lung cancer.Lung cancer remains one of the most deadly types of cancer in the world, making early and accurate diagnosis essential. Damayanti et al. (2023) proposed a hybrid deep learning structure in which folding networks (CNNs) combined with DerSenet to distinguish between malignant and benign lung nodes using radiation imaging. Their model reached an impressive 99.8° campaign. This presented the complementary intensities of CNN properties extraction and Densett in close relations recording subtle lung pathology from CT images. (202They reported over 99.92% on training accuracy, which maintained a robust verification accuracy of 95.6% with an AUC of 0.9967, improving strong generalization skills from Densenet. Complementary studies use the advantages of DENNET 121. The Radioactive Extended Diagnostics Pipeline integrated 93.6% of the specificity of the CT node classification task, 93.6% of hybrid architecture, 93.6% of CNN-CNN security errors, and 93.6% of Dasreng United without shortage. Adenocarcinoma, squamous cell carcinoma and normal tissue classes. Keywords: Deep learning; Lung Cancer Detection; Convolutional Neural Networks (CNNs); Medical Image Analysis ; Early Diagnosis 1. INTRODUCTION Lung cancer remains the driving cause of cancer-related mortality around the world[1], bookkeeping for about one in five cancer passings, generally due to late-stage conclusion and Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 841 https://internationalpubls.com the obtrusive nature of corroborative biopsies. Chest CT filters are the standard for early discovery, however recognizing generous from harmful nodules—even technology-aided—is complex and inclined to misclassification, especially with little injuriesComputed tomography checks are utilized for distinguishing lung illnesses since they offer a intensive see of the tumor within the body and take after its advance[3]. The Creator [2] clarified downsides incorporate that they can't offer assistance with infection discovery at an early organize. The Creator [11] proposed calculation for Lung Maladies Discovery Utilizing Co-Learning from Chest CT Pictures and Clinical Socioeconomics may be a procedure for identifying lung infections in CT check pictures that employments an robotized approach[4],[5]. The Creator [7] utilized an calculation for diagnosing lung ailments is being created utilizing different strategies. The taking after are a few of the limitations: Permits the radiologist to spend more time evaluating the persistent. Executing neural organize frameworks is complex. The Creator [6] proposed a computed tomography filter calculation Picture preparing advances are utilized to analyze the pictures[8]. The passing rate from lung sicknesses is expanding day by day in both youthful and ancient people, and its limits are getting to be more self-evident when compared to other illnesses[9]. The second-request genuine information with respect to spatial association of pixels in a picture is contained in GLCM. The Creator [15] actualized to guarantee the precision of any course of action system; it is more often than not depended on the determination of the foremost suited qualities. As a result, finding a great course of action of highlights, which may be a quantifiable approach that leverages the spatial association of pixels, is basic[10]. 2. LITERATURE REVIEW Cancer Insights, 2021 Siegel RL, Mill operator KD, Fuchs HE, Jemal A (2021) This yearly report from the American Cancer Society gives comprehensive cancer measurements for 2021, counting frequency, mortality, and survival rates over major cancer sorts within the U.S. It emphasizes that whereas in general cancer mortality has declined basically due to progressions in early location and treatment, incongruities endure. The foremost common modern cancer analyze were breast, prostate, and lung cancers, with comparing mortalities intensely affected by cancer sort and arrange. The report underscores a developing predominance of cancers connected to maturing and way of life components, contending for improved anticipation methodologies, progressed screening, and evenhanded get to to care—especially in underserved populaces. . Worldwide Designs of Breast Cancer Rate and Mortality: A Population-Based Cancer Registry Information Examination from 2000 to 2020 Lei S, Zheng R, Zhang S, Wang S, Chen R, Sun K, Zeng H, Zhou J, Wei W (2021) This consider analyzed worldwide patterns in female breast cancer rate and mortality from 2000 to 2015 utilizing GLOBOCAN and CI5plus datasets covering 185 nations. In 2020, an assessed 2.3 million unused cases and 685,000 passings were detailed around the world. Rate shifted essentially, from 112.3 per 100,000 in Belgium to 35.8 in Iran, whereas mortality shifted Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 842 https://internationalpubls.com from 41.0 in Fiji to 6.4 in South Korea. The investigation uncovered rising rate and mortality in China and South Korea, differentiating with decreases in high-income countries just like the USA, UK, and Australia. Key drivers incorporate populace maturing, way of life changes, screening appropriation, and healthcare get to. The discoveries highlight the critical require for focused on methodologies in high-burden districts. Worldwide Breast Cancer Rate, Mortality, and Future Projections Through 2050 Creator Bunch through GLOBOCAN & WHO (2025) An universal consortium counting the IARC and WHO ventures that by 2050, yearly worldwide breast cancer cases will reach around 3.2 million, with passings around 1.1 million—a 38% increment in cases and a 68% rise in mortality from 2022. The consider recognizes maturing, way of life chance components (like liquor utilize and corpulence), and incongruities in healthcare get to as fundamental donors. Most elevated rate remains concentrated in high-income districts such as North America and Western Europe, whereas mortality rates are most noteworthy in low-resource districts like parts of Africa and Polynesia. The creators emphasize fortifying early discovery, extending screening, and tending to disparities to combat 3. PROPOSED SYSTEM We propose a half breed, logical, and edge-ready profound learning framework for lung cancer discovery planned to address impediments of existing approaches. The center of the demonstrate combines a CNN–DenseNet half breed spine, utilizing CNNs for nearby highlight extraction and DenseNet’s thick network for wealthier, progressive representations. Propelled by later works that accomplished 99% exactness with mobile-edge frameworks, our engineering too incorporates transformer-inspired layers for improved setting mindfulness and cancer subtype acknowledgment motivated by Patil et al.’s multimodal cross breed demonstrate .To guarantee vigor and generalizability, we apply an progressed preprocessing pipeline: lung division, normalization, and information enlargement (tint, brightness, differentiate, engineered oversampling). Deduction is optimized for mobile-edge arrangement, empowering computation on-device or cloud-edge to play down idleness and reliance on centralized servers .To make strides clinical acknowledgment, we insert explainability inside the pipeline—integrating Grad CAM-based heatmaps and the DeepXplainer system to outwardly highlight districts impacting expectations and to supply neighborhood XAI elucidations nearby classification comes about At last, the framework incorporates a input circle with edge cloud retraining capabilities, allowing intermittent upgrades utilizing user- corrected or recently collected cases. From cloud-based audit to edge deduction, this secluded system—encompassing division, cross breed classification, explainability, and persistent learning—aims to improve exactness, straightforwardness, and deployability in real-world clinical and resource-constrained situations. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 843 https://internationalpubls.com 4. SYSTEM ARCHITECTURE FIG 5.1 System Architecture a sequential cnn takes ct scan slices as input and passes them through multiple conv layers (filters f, kernels k, strides s), followed by pooling layers to downsample feature maps fig 5.1 5. DATA ANALYSIS AND RESULTS Classification report: The classification report shown represents the performance of a deep learning framework designed for the early detection of lung malignancies. It evaluates four classes: adenocarcinoma, large cell carcinoma, squamous cell carcinoma, and normal tissue. The Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 844 https://internationalpubls.com overall model achieved 82% accuracy, with a macro F1-score of 0.83, indicating strong and balanced performance across all classes. The highest precision and recall were observed for normal cases (1.00 and 0.92), while squamous cell carcinoma had the lowest F1-score (0.71), suggesting room for improvement. These results validate the effectiveness of convolutional neural networks in distinguishing between malignant and non-malignant lung conditions. Confusion matrix: The confusion matrix visualizes the classification performance of a convolutional neural network in identifying lung cancer subtypes and normal tissues. Each row represents the actual class, and each column the predicted class. The model correctly predicted: • 18 adenocarcinoma, misclassifying 2 as large cell carcinoma and 3 as squamous cell. • 18 large cell carcinoma, with minimal confusion (2 misclassified as adenocarcinoma). • 12 normal cases were correctly identified with only one misclassified. • 11 squamous cell carcinoma were accurately detected, though 4 were misclassified. • This indicates strong predictive capability, though some overlap exists among carcinoma types, warranting further refinement. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 845 https://internationalpubls.com Precision-Recall Curve Analysis The precision-recall curve illustrates the performance of a convolutional neural network in predicting various lung cancer types and normal tissue. Each curve represents a class: • Normal tissue (green) achieved perfect precision and recall, reflecting exceptional model confidence and accuracy. • Large cell carcinoma (orange) maintained high precision across increasing recall, indicating reliable classification performance. • Adenocarcinoma (blue) shows moderate variability in both metrics, suggesting a trade- off between false positives and false negatives. • Squamous cell carcinoma (red) exhibited the lowest precision-recall consistency, indicating difficulty in accurate detection. This curve confirms the model's robustness for most classes, while highlighting squamous cell carcinoma as a challenge for improvement. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 846 https://internationalpubls.com ROC Curve Analysis This ROC (Receiver Operating Characteristic) curve evaluates the classification performance of a deep learning model for detecting lung cancer subtypes and normal tissues. Each curve represents the trade-off between the true positive rate and false positive rate for a specific class: • Normal class (green) achieved an AUC of 1.00, indicating perfect classification performance. • Large cell carcinoma (orange) followed closely with an AUC of 0.96, suggesting excellent model reliability. • Adenocarcinoma (blue) had an AUC of 0.92, reflecting strong but slightly less robust performance. • Squamous cell carcinoma (red) had the lowest AUC at 0.89, pointing to moderate effectiveness with room for improvement. Overall, the model demonstrates high discriminative power across all classes, validating its suitability for early and accurate lung cancer detection. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 847 https://internationalpubls.com Accuracy Curve Analysis This accuracy curve visualizes the model's training and validation performance over 43 epochs. The blue line indicates training accuracy, which consistently increases, reaching approximately 97%, signifying effective learning from the dataset. The orange line represents validation accuracy, which shows more fluctuation, stabilizing around 75–82%. The widening gap between training and validation accuracy in later epochs suggests overfitting, where the model performs well on training data but generalizes less effectively to unseen data. This behavior indicates the need for regularization techniques such as dropout, early stopping, or data augmentation to enhance the model’s generalization capability. Loss Curve Evaluation The loss curve presented here illustrates the training and validation loss over 43 epochs during the learning process of a deep learning model for lung cancer detection. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 848 https://internationalpubls.com • The blue line represents training loss, which steadily decreases, reaching below 0.2, indicating that the model is effectively minimizing error on the training set. • The orange line shows validation loss, which also decreases but with more fluctuations, stabilizing around 0.6. The consistent gap between training and validation loss suggests overfitting, where the model learns the training data well but performs less effectively on unseen data. Applying regularization strategies such as dropout, batch normalization, or early stopping may help improve generalization. 6. CONCLUSIONS AND FUTURE WORK Conclusion In this consider, we shown an advanced lung cancer disclosure framework utilizing a half breed Convolutional Neural Organize (CNN) moved forward by DenseNet modules. The system outlined momentous capability, classifying liberal and hurtful pneumonic handles with an accuracy of 99.48%, outflanking earlier approaches like mRFCN’s ~97% accuracy . This alters with creating demonstrate that half breed CNN–DenseNet models give energetic classification execution in lung imaging errands. Other than, afterward gathering techniques uniting DenseNet and CNN—especially when combined with edge computing and data fusion—have fulfilled ≥ 99% precision in real-world CT channel appraisals, highlighting the reasonable potential of these systems.These comes approximately emphasize the achievability of sending CNN–DenseNet cross breeds in clinical and resource-limited circumstances, publicizing tall accuracy, moo inertia, and real-time finding, while tending to interpretability through strategies like Grad-CAM. Looking ahead, coordination explainability, mobile-edge course of action, and discontinuous appear retraining will progress move forward system determination and reasonability. In the long run, this explore contributes a high-performance, clear, and deployable expressive gadget arranged to back early revelation and treatment of lung cancer, promising basic influence on understanding comes about around the world. Future scope: Looking ahead, the scene of AI-driven lung cancer discovery is balanced for transformative development. One effective course is multi-omics integration, where AI frameworks combine radiological imaging with genomic, transcriptomic, and proteomic information to provide a comprehensive, personalized symptomatic device. For occurrence, radiogenomic approaches—linking CT picture highlights with EGFR transformation status—show guarantee in anticipating treatment reaction and guess Furthermore, endeavors in AI-based pathology utilizing computerized histology and cytology pictures are developing to move forward symptomatic precision by diminishing untrue negatives and upgrading cellular-level bits of knowledge Another key progression includes intraoperative AI bolster, such as real-time tumor boundary depiction and expanded reality visualization in thoracic surgery—achieving exact tumor Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 3s (2025) 849 https://internationalpubls.com evacuation with minimized healthy-tissue affect This adjusts with the broader drift of AI- enhanced surgical arranging and route coordination mechanical frameworks and 3D visualization instruments REFERENCES 1) R. L. Siegel, K. D. Miller, H. E. Fuchs, and A. Jemal, “Cancer Statistics, 2021.,” CA. Cancer J. Clin., vol. 71, no. 1, pp. 7–33, Jan. 2021, doi: 10.3322/caac.21654. 2) S. Lei et al., “Global patterns of breast cancer incidence and mortality: A population‐based cancer registry data analysis from 2000 to 2020,” Cancer Commun., vol. 41, no. 11, pp. 1183–1194, 2021. 3) M. M. Cackowski et al., “The absence of lymph nodes removed (pNx status) impacts survival in patients with lung cancer treated surgically,” Surg. Oncol., vol. 48, p. 101941, 2023. 4) S. S. Raoof, M. A. Jabbar, and S. A. Fathima, “Lung Cancer Prediction using Machine Learning: A Comprehensive Approach,” in 2020 2nd International Conference on Innovative Mechanisms for Industry Applications (ICIMIA), 2020, pp. 108–115. doi: 10.1109/ICIMIA48430.2020.9074947. 5) Y. Chen, E. Zitello, R. Guo, and Y. Deng, “The function of LncRNAs and their role in the prediction, diagnosis, and prognosis of lung cancer,” Clin. Transl. Med., vol. 11, no. 4, p. e367, 2021. 6) J. A. Barta, C. A. Powell, and J. P. Wisnivesky, “Global epidemiology of lung cancer,” Ann. Glob. Heal., vol. 85, no. 1, 2019. 7) M. A. Heuvelmans et al., “Lung cancer prediction by Deep Learning to identify benign lung nodules,” Lung Cancer, vol. 154, no. November 2020, pp. 1–4, 2021, doi: 10.1016/j.lungcan.2021.01.027. 8) J. Jumanto, M. F. Mardiansyah, R. Pratama, M. F. Al Hakim, and B. Rawat, “Optimization of breast cancer classification using feature selection on neural network,” J. Soft Comput. Explor., vol. 3, no. 2, pp. 105–110, 2022, doi: 10.52465/joscex.v3i2.78. 9) B. He et al., “Image segmentation algorithm of lung cancer based on neural network model,” Expert Syst., vol. 39, no. 3, p. e12822, 2022. 10) Q. Wang, Y. Zhou, W. Ding, Z. Zhang, K. Muhammad, and Z. Cao, “Random forest with self-paced bootstrap learning in lung cancer prognosis,” ACM Trans. Multimed. Comput. Commun. Appl., vol. 16, no. 1s, pp. 1–12, 2020. 11) S. Wankhade and S. Vigneshwari, “A novel hybrid deep learning method for early detection of lung cancer using neural networks,” Healthc. Anal., p. 100195, 2023.