A Symptom-Aware Deep Learning Model for Lung Cancer Detection Using Sparse Channel Attention and Sequential CT Analysis

Abstract

Lung cancer detection is challenging due to late diagnosis, complex CT image patterns, and limitations in existing deep learning models such as sensitivity to noise, redundant feature extraction, and lack of clinical context. To address these issues, this study proposes a symptom-aware deep learning framework that integrates ASNS-Net for adaptive noise suppression, SCAS-Net for efficient feature selection, and MVC-BLDNet (CNN with Bi-LSTM) for capturing both spatial and inter-slice sequential features. By incorporating patient symptom information and advanced attention mechanisms, the model enhances feature representation and reduces false predictions. The framework also improves computational efficiency by eliminating irrelevant features and focusing on diagnostically significant regions. Furthermore, the integration of sequential learning enables better understanding of tumour progression across CT slices. Overall, the proposed approach improves accuracy, robustness, and generalization, making it effective for early and reliable lung cancer detection and supporting clinical decision-making.

Introduction

The document proposes a symptom-aware deep learning framework for lung cancer detection using CT scan images, aimed at improving early diagnosis, accuracy, and reliability in computer-aided medical imaging.

Summary

Lung cancer is difficult to detect early, and traditional methods struggle with noise in CT images, redundant feature extraction, and lack of clinical context. While CNN-based deep learning models have improved detection, they still face limitations in accuracy, interpretability, and handling sequential CT data.

To address this, the proposed system introduces a three-stage deep learning pipeline:

• ASNS-Net (Adaptive Symptom-Aware Noise Suppression Network): Removes noise from CT images while preserving tumor-relevant regions using patient symptom guidance.
• SCAS-Net (Sparse Channel Attention Selection Network): Extracts meaningful features and removes redundant ones using attention mechanisms.
• MVC-BLDNet (Multi-View CNN + Bi-directional LSTM): Combines spatial feature learning from multiple CT views with sequential slice analysis to capture inter-scan dependencies.

The system architecture includes preprocessing (normalization, resizing, denoising), deep feature extraction, classification, and explainability using Grad-CAM, which highlights tumor regions for transparency. It also provides prediction confidence scores and a user-friendly interface for clinical use.

The literature review shows that existing methods (DenseNet, attention CNNs, segmentation models, and hybrid 2D/3D networks) improve performance but suffer from issues like high computational cost, limited generalization, and poor temporal modeling.

Results

The proposed system achieves:

• High classification accuracy for cancer vs. normal CT scans
• Improved robustness through noise reduction and attention-based feature selection
• Better tumor localization validated with ground truth masks
• Enhanced interpretability using explainable AI (Grad-CAM)

Conclusion

The proposed lung cancer detection system successfully integrates advanced deep learning techniques to provide accurate and efficient diagnosis using CT scan images. By combining models such as ASNS-Net, SCAS-Net, and MVC-BLDNet, the system effectively enhances image quality, extracts meaningful features, and performs reliable classification of cancerous and normal cases. The inclusion of Explainable AI techniques like Grad-CAM improves transparency by highlighting important regions influencing the prediction, making the system more trustworthy for medical professionals. The user-friendly GUI further enhances usability by allowing easy image upload, visualization of results, and comparison with ground truth masks. Overall, the system demonstrates strong performance in terms of accuracy, reliability, and interpretability, making it a valuable tool for early lung cancer detection. It has the potential to assist radiologists in clinical decision-making and contribute to improved patient outcomes through timely diagnosis.

References

Xi, Zhichao, Rongchen Dai, Yufei Ze, Xue Jiang, Mengfan Liu, and Hongxi Xu. \"Traditional Chinese medicine in lung cancer treatment.\" Molecular cancer 24, no. 1 (2025): 57. [2] Uddin, Jia. \"Attention-based densenet for lung cancer classification using CT scan and histopathological images.\" Designs 8.2 (2024): 27. [3] Haque, Fariha, et al. \"An End-to-End Concatenated CNN Attention Model for the Classification of Lung Cancer With XAI Techniques.\" IEEE Access (2025). [4] Wang, Guotai, et al. \"Semi-supervised segmentation of radiation-induced pulmonary fibrosis from lung CT scans with multi-scale guided dense attention.\" IEEE transactions on medical imaging 41.3 (2021): 531-542. [5] Said, Yahia, et al. \"Medical images segmentation for lung cancer diagnosis based on deep learning architectures.\" Diagnostics 13.3 (2023): 546. [6] Li, Yongbin, et al. \"Lung nodule detection using a multi-scale convolutional neural network and global channel spatial attention mechanisms.\" Scientific Reports 15.1 (2025): 12313. [7] Deepa, J., et al. \"A novel lung cancer diagnosis model using hybrid convolution (2D/3D)-based adaptive DenseUnet with attention mechanism.\" Network: Computation in Neural Systems (2025): 1-58. [8] .Zhao, Shen, Hongyun Zhao, Weiwei Yang, and Li Zhang. \"The next generation of immunotherapies for lung cancers.\" Nature reviews Clinical oncology 22, no. 8 (2025): 592-616. [9] Mountzios, Giannis, Longhua Sun, Byoung Chul Cho, Umut Demirci, Sofia Baka, Mahmut Gümü?, Antonio Lugini et al. \"Tarlatamab in small-cell lung cancer after platinum-based chemotherapy.\" New England Journal of Medicine 393, no. 4 (2025): 349-361. [10] Su, Po-Lan, Naoki Furuya, Alahmadi Asrar, Christian Rolfo, Zihai Li, David P. Carbone, and Kai He. \"Recent advances in therapeutic strategies for non-small cell lung cancer.\" Journal of hematology & oncology 18, no. 1 (2025): 35.

Copyright

Copyright © 2026 Aswitha V. V., Dhakshanya A, Dhuraganandhini K. R., Janani S. A., Mr. Althaf Ahamed S. A.. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.