Virtual Library
Start Your Search
N. Aramaki
Author of
-
+
P1.07 - Immunology and Immunotherapy (ID 693)
- Event: WCLC 2017
- Type: Poster Session with Presenters Present
- Track: Immunology and Immunotherapy
- Presentations: 1
- Moderators:
- Coordinates: 10/16/2017, 09:30 - 16:00, Exhibit Hall (Hall B + C)
-
+
P1.07-012 - Prediction Sensitivity of PD-1 Checkpoint Blockade Using Pathological Tissues Specimens by Novel Computerized Analysis System (ID 8851)
09:30 - 09:30 | Author(s): N. Aramaki
- Abstract
Background:
Recent development of immune checkpoint blockade such as anti-PD-1 antibody brought great benefits to non-small cell lung cancer (NSCLC) patients. However, some population of NSCLC showed resistance and pseudo-progressions against anti-PD-1 checkpoint blockade. Thus, it is very important for developing biomarkers which predict of efficacy of PD-1 checkpoint blockade. In this background, we developed novel digital pathology system that predict for response to anti-PD-1 checkpoint blockade using H&E staining sections and technology of AI.
Method:
In this study, we extract 361 ROIs(Region of Interest) and 254,205 nuclei were measured from NSCLC cases that treated with anti-PD-1 antibody. We used ilastik for nuclei image segmentation, CellProfiler and our CFLCM tool for features measurement, 992 features are evaluated for each ROI. At first, we analyzed by step-wise discriminant analysis for select the effective features, and using canonical discriminant analysis and SVM (Support vector Machine) RBF kernel model discrimination, we analyzed morphological data based PD-1 blockade response on statistical platform R.
Result:
Except undeterminable cases, we got the more than 95% accuracy level discrimination results. The mapping the discriminant scores, SD cases were mapped in the middle of PR and PD. Only using the average and standard deviation of ROIs’ nuclei shape features (size, roundness, perimeter, etc.) and inside nuclei features (mainly chromatin texture) more than 90% discrimination results were obtained. This means the nuclei morphological data is more important than CFLCM (pleomorphism and heterogeneity measurement data). We challenged the prediction for undeterminable cases by using canonical discriminant and SVM.
Conclusion:
This time analysis is small number samples, so the results application robustness may be limited. But our results show the possibility for clinical response prediction even on the pre-treatment pathological tissues specimens.
