Start Your Search
MA15 - Usage of Computer and Molecular Analysis in Treatment Selection and Disease Prognostication (ID 141)
- Event: WCLC 2019
- Type: Mini Oral Session
- Track: Pathology
- Presentations: 1
- Now Available
MA15.02 - Deep Learning Approach for Automated Tumor Cells Detection and Estimation of PD-L1 22C3 Assay Expression in Lung Adenocarcinoma (Now Available) (ID 577)
15:45 - 17:15 | Author(s): Shaoping Ling
It is vital and challenging to assess an accurate PD-L1 expression status on tumor cells for immunotherapy in lung cancer. The purpose of this study was to set up an automated system to detect the tumor cells and estimate the tumor proportion score (TPS) of PD-L1 immunohistochemistry (IHC) expression for lung adenocarcinoma based on deep learning, and provide a potential Artificial Intelligences (AI) assistive diagnostic tool in the quantification of PD-L1 interpretation.Method
Fifty PD-L1 22C3 IHC slides of lung adenocarcinoma samples on digitized whole-slide images (WSI) database was employed. We first designed a model with a fully convolutional neural network (FCNN) based on U-ResNet architecture to obtain the cancer segmentation. Representative regions were selected from each slide, and 100 regions were collected for manual annotations as a training set for cancer detection. Another 50 regions were used to validate the performance of automated cancer detection and TPS estimation as a test set. After the quality control, a whole model of automated cancer cell segmentation and membrane positive estimation was set up on standard PD-L1 22C3 IHC staining. TPS could be automatically predicted by AI tool and then compared with the interpretations of pathologists.Result
The results of automated lung adenocarcinoma cells segmentation on the test set of 22C3 IHC staining showed a moderate sensitivity (71.46%) with a high specificity (95.94%) which was much more crucial for TPS counting. In rest 43 out of 50 regions after a quality control, TPS estimated by the automated PD-L1 analysis based on cancer segmentation showed a significant correlation with the average scores (r=0.9609, p<0.001) and the median scores (r=0.9523, p<0.001) of pathologists' interpretations.
We provide an automated tumor cells detection and TPS estimation model for lung adenocarcinoma and demonstrate the potential of using machine learning methods to access PD-L1 IHC status conveniently. A further validation of AI tool for automated scoring PD-L1 in diagnostic routine is highly recommended in the future.
Only Active Members that have purchased this event or have registered via an access code will be able to view this content. To view this presentation, please login or select "Add to Cart" and proceed to checkout.