To develop a deep learning algorithm for accurate screening of ALK alterations specifically in non-small cell lung cancer (NSCLC) from H&E-stained pathological images.
Key Findings:
The proposed model achieved over 95% accuracy with both resection and biopsy datasets, indicating its reliability.
DeepPATHO can potentially reduce unnecessary medical costs associated with current screening methods and improve the understanding of genetic alterations associated with pathological phenotypes.
A publicly available Python-based open-source software package was developed for clinical application, facilitating wider adoption.
Interpretation:
The high accuracy of the DeepPATHO model suggests it can effectively screen for ALK alterations in NSCLC, which may significantly enhance clinical decision-making and patient management.
Limitations:
The model's performance needs to be validated across larger and more diverse cohorts, including various demographics and stages of NSCLC.
The black-box nature of deep learning models can hinder trust in clinical applications, necessitating transparency in model decision-making.
Conclusion:
Deep learning presents a promising approach for enhancing the screening of genetic alterations in lung cancer, potentially leading to better-targeted therapies and improved patient outcomes.
