ساخت اکانت کاربری رایگان

مقالات پذیرفته شده کنگره

  • صفحه نخست
  • مقالات پذیرفته شده اولین کنگره بین المللی ژنتیک سرطان
بازنشر

  • Application of Artificial Intelligence in the Diagnosis of Lung Cancer from Pathology Slides: A Systematic Review

  • Mohammad Ali Nazmabadi Nezhad,1,* Parichehr Ebrahimi Shahabadi,2
    1. Doctorate of Veterinary Medicine Graduated, Faculty of Veterinary Medicine, Shahid Bahonar University of Kerman, Kerman, Iran
    2. Doctorate of Veterinary Medicine Graduated, Faculty of Veterinary Medicine, Shahid Bahonar University of Kerman, Kerman, Iran


  • Introduction: Lung cancer is the second most common cancer and the leading cause of cancer-related death in the world, and about 85% of cases are non-small cell lung cancer (NSCLC). Accurate and early diagnosis of this disease plays a vital role in prognosis and selection of treatment strategy. In recent years, artificial intelligence, especially deep learning, has made a significant transformation in digital pathology. Whole Slide Images, which convert microscopic tissue slides into high-quality digital images, have provided a suitable platform for applying artificial intelligence algorithms. This systematic review aims to review and synthesize the existing evidence on the application of artificial intelligence algorithms in the diagnosis of lung cancer through the analysis of pathology slides, with an emphasis on various diagnostic tasks, including classification of malignant versus benign tissue, determination of tumor growth patterns, histological subtyping, and prediction of molecular markers.
  • Methods: This systematic review was conducted according to the PRISMA guidelines. A comprehensive search of PubMed, Scopus, Web of Science, Cochrane, and Embase databases was conducted for articles published between January 2020 and April 2026. Key keywords included a combination of "Artificial Intelligence", "Deep Learning", "Whole Slide Image", "Histopathology", "Lung Cancer", "EGFR", "NSCLC", and "Digital Pathology". Inclusion criteria included original studies that used artificial intelligence algorithms (machine learning or deep learning) to analyze hematoxylin-eosin (H&E)-stained pathology images in the diagnosis or prediction of lung cancer features. Reviews, case reports, and studies lacking external validation or extractable performance data were excluded from the analysis.
  • Results: Analysis of the 26 included studies showed that AI algorithms have been used in lung cancer pathology in several major areas. In the area of classification and diagnosis, deep learning models have been able to distinguish malignant from non-malignant tissue with high accuracy. According to the findings of a comprehensive meta-analysis, AI in lung cancer imaging diagnosis has shown a combined sensitivity of 0.86, a combined specificity of 0.86, and an area under the curve (AUC) of 0.92. Regarding histological subtyping, models have performed very well in differentiating adenocarcinoma from squamous cell carcinoma. The most advanced and promising application of AI in this area is the prediction of molecular markers and gene mutations directly from H&E images. Deep learning algorithms have particularly demonstrated outstanding performance in three areas: classification of malignancy versus benignity with an AUC of approximately 0.92, histological subtyping with an AUC of up to 0.999, and noninvasive prediction of key molecular mutations, including ALK (84% sensitivity, 85% specificity) and EGFR (80% sensitivity, 77% specificity) directly from H&E slides. Despite these promising results, significant challenges were identified. The most important challenge is the lack of robust external validation. A scoping review found that of the available studies, many used limited and unrepresentative datasets, and the studies were mostly retrospective and case-control designs. Methodological heterogeneity in image preparation, different patch sizes, and lack of standardized protocols are major barriers to model generalizability. Also, the sample sizes of many studies are small, and only a limited number of models have been validated in ethnically and geographically diverse populations.
  • Conclusion: This systematic review demonstrates that artificial intelligence, with its ability to extract subtle patterns not visible to humans from pathology images, has great potential to revolutionize lung cancer diagnosis. However, the lack of external validation in diverse populations, methodological heterogeneity, and small sample sizes are major barriers to the introduction of these technologies into routine clinical use. Future steps should focus on standardizing protocols, conducting prospective multicenter studies with large and diverse datasets, and designing randomized clinical trials to compare the performance of intelligence models.
  • Keywords: Artificial intelligence, lung cancer, full-image slide, digital pathology

به خانواده بزرگ کنسر ژنتیکس و ژنومیکس سرطان بپیوندید!

ایجاد حساب کاربری رایگان