Oke Dimas Asmara

M.D.

University of Groningen

PhD Researcher

NLST (Learn more about this study)

NLST-1272

Jun 17, 2024

SynapticEdge: Deep learning-based diagnostic tool for lung cancer pathology images

Lung cancer has the highest incidence and mortality rate among all cancers globally, accounting for over 2.4 million new cases and over 1.8 million deaths worldwide annually. Non-small cell lung cancer (NSCLC) represents over 80% of lung cancer cases, with adenocarcinoma being the most common histologic subtype. Identifying its specific type holds a significant role in the management, with histopathology serving as the gold standard. Histopathology specimens will subsequently be used for immunohistochemical and molecular gene mutation tests that guide personalized therapy. However, as the most frequent and deadliest cancer, confirming diagnostic for specific lung cancer type must be performed as quickly as possible to avoid delaying initial treatment. The high workload of pathologists and the interobserver variability are challenges in achieving accurate diagnosis promptly. The time-consuming process and the limited availability of facilities, especially in low- to middle-income countries, can severely hinder the diagnostic process. To overcome those problems, artificial intelligence (AI)-based approach for recognizing cancer cell in lung tissue samples may offer a solution.

Recognizing cancer patterns in a pathology image is important, and computer vision with deep learning approach has proven effective in pattern recognition. This has been demonstrated in multiple previous studies across various types of cancer including breast, colon cancer and lung cancer. With rapid advancements in deep learning technology and the emergence of digital pathology in the last decade, there has never been a better time to develop AI-based tool for detecting lung cancer from pathology images. We are planning to develop an AI model to detect lung cancer in pathology images using high-quality training datasets from different sources. The addition of the National Lung Screening Trial (NLST) dataset to our training set will greatly improve the accuracy of the model. The model will be further refined through external validation with real-world datasets from diverse population, including European and Asian demographic. By taking this important step, we hope to facilitate the integration of AI technology into clinical pathway within the medical field.