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

Abstract

Efficient and accurate identification of genetic alterations of non-small cell lung cancer is a critical diagnostic process for targeted therapies. Utilizing advanced modern deep learning is a potential solution that can accurately predict genetic alterations from H&E-stained pathological images without additional testing procedures and costs. However, clinically applicable predictive power for Anaplastic Lymphoma Kinase (ALK) rearrangement has yet to succeed. To tackle these issues, we have developed a pathologically interpretable, evidence-based deep learning algorithm to screen ALK alterations to reduce unnecessary medical costs and understand the association between genetic alterations and pathological phenotypes. The proposed model resulted in +95% accuracy with both resection and biopsy datasets, which can be applicable in the clinic. The deep learning approach can maximize the benefits for screening genetic alterations as well as provide the most clinical utility. A stand-alone Python-based open-source software package is publicly available.