2023 World Conference on Lung Cancer (Posters)-P1.22. Immediate Biomarkers Profiling of Actionable NSCLC Panel from a Digitized H&E Biopsy Image

Pdf Summary

This session at the WCLC 2023 conference focused on the use of image-based biomarkers profiling in lung cancer pathology. Non-Small Cell Lung Cancer (NSCLC) is the most common type of lung cancer, and there are major challenges in providing optimal personalized treatment due to tissue shortage, long reporting times, and bureaucracy processes. Researchers presented the use of an image-based biomarkers profiling solution to address these issues.<br /><br />The study involved analyzing de-identified digitized H&E slides using an image-based lung cancer profiling solution. They examined eight actionable biomarkers, including gene fusions and actionable alterations, in a cohort of 1914 NSCLC patients. The samples were collected from multiple sites and included biopsies and resections. The researchers used a self-supervision model with dynamic data augmentation and Multiple Instance Learning algorithms, training the digital slides from various public and commercial databases.<br /><br />Comparing the results from the image-based biomarkers profiling to the official molecular results, the researchers found an average sensitivity and specificity of 92.4% and an average accuracy of 92.6% when focusing on high-confidence cases. The Average Area Under the Curve (AUC) was 0.934, indicating the effectiveness of the approach. The study also highlighted the performance of specific biomarkers, such as NTRK-fusions, with 94% accuracy and an AUC of 0.972.<br /><br />In conclusion, the research demonstrated the potential of using image-based analysis and deep learning algorithms for immediate profiling of lung cancer biomarkers. This approach offers a generalized solution that can overcome disparities in imaging devices, population backgrounds, and staining protocols. It provides benefits such as routine profiling for all NSCLC patients in both early and advanced stages, as well as optimizing the pathological diagnosis process. The results highlight the value of incorporating immediate image-based molecular profiling into routine pathological workflows.

Asset Subtitle

Nurit Paz-Yaacov

Meta Tag

Speaker Nurit Paz-Yaacov

Topic Pathology & Biomarkers: Genetic Biomarkers

Keywords

WCLC 2023 conference

image-based biomarkers profiling

lung cancer pathology

Non-Small Cell Lung Cancer (NSCLC)

personalized treatment

tissue shortage

long reporting times

bureaucracy processes

gene fusions

actionable alterations