基于激光诱导击穿光谱结合深度学习的肺部肿瘤诊断方法研究

doi:10.3964/j.issn.1000-0593(2025)10-2730-07

Abstract
References
Related Citation (15)

Download: PDF (6743 KB)
Export: BibTeX | EndNote (RIS)

Abstract Lung cancer is the most deadly form of cancer worldwide, with a high morbidity and mortality rate. The accuracy of a patient's diagnosis directly impacts their treatment plan and the likelihood of recurrence post-surgery. Conventional diagnostic methods are often dependent on the subjective assessment of medical professionals and are time-consuming. Consequently, there is an urgent need for a lung tumor diagnostic method that can provide objective and quantitative metrics, facilitating rapid detection. The objective of this study is to assess the viability of a novel approach that integrates laser-induced breakdown spectroscopy (LIBS) with deep learning network models for the expeditious and in situ diagnosis of diseased lung tissues. The LIBS technique was employed to quantitatively analyze the elemental composition of cancerous tissues, lung tumors, and normal tissues from 45 patients. This analysis enabled the rapid detection of lung tumors and the acquisition of elemental differences between diseased and normal tissues. A total of 12 characteristic spectral lines of 6 elements, including Ca, Na, K, etc., were selected as inputs to the model through multivariate analysis. To address the challenges posed by the intricate preprocessing and feature extraction in tumor and normal tissue spectral data, a deep learning spectral feature processing system with ResNet18 as the primary network was developed. In conjunction with this system, 3 machine learning models were designed to be utilized in conjunction with the Random Forest feature extraction method. To facilitate a comprehensive comparison, a quadratic recursive machine learning spectral feature processing system was also established. The findings indicate that the deep learning network model exhibits a substantially superior recognition capability compared to the other three machine learning models. Its accuracy, sensitivity, and specificity reach 99.6%, 100%, and 99.3%, respectively. The model demonstrated a 99.6% accuracy, 100% precision, and 99.3% recall in the recognition of spectral data from 140 tumor tissues and 139 normal tissues. The model's balanced ability to recognize different spectral data types while ensuring recognition accuracy, in conjunction with its demonstrated generalization and robustness, is noteworthy. The study above demonstrates that, in the context of high-dimensional and abstract spectral feature information in tumors and normal tissues, the convolution and pooling regions in the deep learning network model exhibit a superior capacity to extract nonlinear features in comparison to traditional chemometric methods. This enhanced capability enables the expeditious identification of similarity and difference information in tumors and normal tissues. The integration of LIBS with deep learning has been demonstrated to facilitate the acquisition of objective, quantitative data concerning diseased tissue in the context of lung cancer diagnosis. This approach has been demonstrated to provide a rapid, precise, and robust method for identifying lung tumors.

Key words：Laser induced breakdown spectroscopy; Rapid diagnosis of lung tumors; Deep learning

Received: 2025-01-16 Accepted: 2025-06-02

ZTFLH:

O433.4

Corresponding Authors: FANG Jian E-mail: 1419404982@qq.com

Cite this article:

SUN Hao-ran,ZHAO Chun-yuan,LIN Xiao-mei, et al. Research on Lung Tumor Diagnosis Method Based on Laser-Induced Breakdown Spectroscopy Combined With Deep Learning[J]. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2025, 45(10): 2730-2736.

URL:

https://www.gpxygpfx.com/EN/10.3964/j.issn.1000-0593(2025)10-2730-07 OR https://www.gpxygpfx.com/EN/Y2025/V45/I10/2730

[1] Suenghataiphorn T, Srikulmontri T, Kulthamrongsri N, et al. Journal of Clinical Oncology, 2024, 42(23): 212.
[2] Zhong S, Borlak J. Archives of Toxicology, 2024, 98(3): 957.
[3] Kang M S, Chae W R, Lee Y J, et al. Toxics, 2024, 12(1): 20.
[4] Guo T, Chen S, Wang Y, et al. International Journal of Cancer, 2024, 154(2): 251.
[5] Kayyal-Tarabeia I, Zick A, Kloog I, et al. International Journal of Epidemiology, 2024, 53(4): dyae093.
[6] Liu F, Si C, Chen L, et al. Molecular Nutrition & Food Research, 2024, 68(19). doi: 10.1002/mnfr.202400448.
[7] Yao L, Zhou H Y, Cai L, et al. Journal of Clinical Oncology, 2024, 42(16): 1563.
[8] Hanaoka T, Matoba H, Nakayama K O M. Radiological Physics and Technology, 2024, 17(1): 71.
[9] Li W, Huang J, Shen C, et al. Acta Pharmaceutica Sinica B, 2024, 14(2): 751.
[10] Rani V J, Thanammal K K. Wireless Personal Communications, 2023, 132(2): 1453.
[11] Kimura T, Akazawa T, Mizote Y, et al. International Journal of Clinical Oncology, 2024，29(6): 771.
[12] Tominaga M, Yamazaki M, Umezu H, et al. Academic Radiology, 2024，31(9): 3801.
[13] NIU Guang-hui, ZHANG Ye-jian, LIN Qing-yu(牛广辉，张业建，林庆宇). Chinese Journal of Inorganic Analytical Chemistry(中国无机分析化学), 2024, 14(2): 168.
[14] Tankova V, Atanassova V, Pirovska A, et al. Optical & Quantum Electronics, 2024, 56(7): 1193.
[15] Manelski H T, Wiens R C, Bousquet B, et al. Spectrochimica Acta Part B: Atomic Spectroscopy, 2024, 222: 107061.
[16] Kiss K, Sindelarova A, Krbal L, et al. Journal of Analytical Atomic Spectrometry, 2021, 36(5): 909.
[17] Teng G, Wang Q, Zhang H, et al. Spectrochimica Acta Part B: Atomic Spectroscopy, 2020, 165：105787.
[18] Chen X, Li X, Yu X, et al. Spectrochimica Acta Part B: Atomic Spectroscopy, 2018, 139: 63.
[19] Chu Y, Chen T, Chen F, et al. Journal of Analytical Atomic Spectrometry, 2018, 33(12): 2083.
[20] Askr H, El-Dosuky M, Darwish A, et al. Applied Soft Computing, 2024, 164: 112009.