Evidential deep learning-based ALK-expression screening using H&E-stained histopathological images.

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Accurate identification of genetic alterations in non-small cell lung cancer (NSCLC) is crucial for effective targeted therapies.
• Current methods for detecting genetic alterations like Anaplastic Lymphoma Kinase (ALK) rearrangement can be costly and time-consuming.
• Deep learning offers a potential solution for predicting genetic alterations directly from standard histopathological images.

Purpose of the Study:

• To develop and validate a deep learning algorithm for the pathological interpretation and prediction of ALK rearrangements in NSCLC.
• To assess the clinical applicability and accuracy of the AI model in screening ALK alterations.
• To reduce unnecessary medical costs associated with genetic testing and explore genotype-phenotype associations.

Main Methods:

• Development of a pathologically interpretable, evidence-based deep learning algorithm.
• Training and validation of the model using H&E-stained pathological images from NSCLC resection and biopsy specimens.
• Evaluation of the model's predictive accuracy and clinical utility.

Main Results:

• The deep learning model achieved over 95% accuracy in predicting ALK alterations on both resection and biopsy datasets.
• The developed algorithm demonstrates high potential for clinical application in NSCLC diagnostics.
• The study provides insights into the association between genetic alterations and pathological phenotypes.

Conclusions:

• Deep learning provides an accurate and efficient method for screening ALK alterations in NSCLC, potentially reducing healthcare costs.
• The AI-driven approach offers significant clinical utility for guiding targeted therapy selection.
• A publicly available, open-source Python software package facilitates the implementation of this technology.