PV22 PRESENTER

 

Lami HeadshotKris Lami, MD

Graduate student

Nagasaki University

 

 

 

 

 

 

 

 

Presenting

Standardized clasification of lung adenocarcinoma sybtypes and improvement of gradin assessment using deep learning algorithm

 

Abstract

Background: The histopathological distinction of lung adenocarcinoma (LADC) subtypes is subject to a high inter-observer variability which can compromise the optimal assessment of patients' prognosis. Therefore, this study focused on the development of convolutional neural networks (CNNs) capable of distinguishing LADC subtypes and predicting overall survival according to the LADC tumor grades recently established by the International Association for the Study of Lung Cancer pathology committee.

 

Methods: To minimize the influence of inter-observer variability, ground truth images for different LADC subtypes were obtained from a consensus of international expert pulmonary pathologists, using a clustering approach. Two clusters were obtained with their own ground truth images, and two different models (AI-1 and AI-2) were trained with the EfficientNet b3 architecture to predict 8 different classes (lepidic, acinar, papillary, micropapillary, solid, invasive mucinous adenocarcinoma, other carcinoma types, and no carcinoma cells). The two trained models have been tested on an independent cohort of 133 patients.

 

Results: The models achieved high precision, recall, and F1-score for the majority of classes, exceeding 0.90. Clear stratification of the three LADC grades has been obtained in the prediction of the overall survival by the two models. Moreover, one of the trained models had a more significant grading prediction than 14 out of the 15 pulmonary pathologists (p = 0.0003). Both trained models showed high stability in the segmentation of every pair of predicted grades with low variation of hazard ratio across 200 bootstrapped samples.

 

Conclusions: These findings showed that trained CNNs improved pathologists' diagnostic accuracy, standardized LADC subtypes recognition, and refined LADC grades assessment. Trained models are promising tools for the assistance in the routine evaluation of LADC grades in clinical practice.

 

Objectives

  1. Get to know about the development of convolutional neural networks capable of accurately classifying and grading lung adenocarcinoma on a whole-slide image.

 

Biography

Kris Lami, a Congolese citizen from the DRC, got his MD degree at the Université Prostestante au Congo, in Kinshasa. He got his Ph.D. in Medical Science in 2022, in the Department of Pathology at Nagasaki University, Japan. Currently Assistant Professor in the Department of Pathology Informatics at the Graduate School of Biomedical Sciences of Nagasaki University, he is involved in daily surgical pathology diagnoses of cases from several hospitals and institutions across Japan using digital pathology and whole slide images. He has participated in several international seminars including USCAP, and the annual meeting of the Japanese Society of Pathology, among others. His current field of research is mainly the recognition of lung adenocarcinoma subtypes using deep learning algorithms.

 


40+ Expert Presenters

View the PV22 agenda.

CLICK HERE FOR MORE INFO

Register for #PathVisions22! 

October 16-18 | Las Vegas, NV

CLICK HERE FOR MORE INFO

Support & Exhibit Opportunities

Trade Show Sold Out!

CLICK HERE FOR MORE INFO