A Computational Tumor-Infiltrating Lymphocyte Assessment Method Comparable with Visual Reporting Guidelines for
Peng Sun1, Jiehua He1, Xue Chao1
1State Key Laboratory of Oncology in South China, Collaborative Innovation Center for Cancer Medicine, Guangzhou, P. R. China; Department of Pathology, Sun Yat-sen University Cancer Center, Guangzhou, P. R. China.
Ebiomedicine
|July 19, 2021
Summary
A new computational method for assessing tumor-infiltrating lymphocytes (TILs) in triple-negative breast cancer (TNBC) shows prognostic value. This deep learning approach, computational TIL assessment (CTA), complements visual assessment for improved clinical decision-making.
Area of Science:
- Oncology
- Computational Pathology
- Biomedical Imaging
Background:
- Tumor-infiltrating lymphocytes (TILs) are crucial biomarkers in triple-negative breast cancer (TNBC).
- Existing visual TILs assessment (VTA) methods have limitations.
- A deep learning-based computational TIL assessment (CTA) was developed to address these limitations.
Purpose of the Study:
- To establish and validate a deep learning-based CTA method for TNBC.
- To compare CTA with VTA for scoring concordance and prognostic value.
- To determine a practical CTA workflow for clinical integration.
Main Methods:
- Three deep neural networks were trained for nuclei segmentation, classification, and necrosis classification.
- An automatic TIL (aTIL) score was generated and compared with manual TIL (mTIL) scores from pathologists.
- The study included Asian (n=184) and Caucasian (n=117) TNBC cohorts.
Main Results:
- Intraclass correlations between aTILs and mTILs ranged from 0.40 to 0.70.
- The aTIL score demonstrated prognostic value for disease-free survival (DFS) in both cohorts.
- A composite model integrating manual and computational TIL scores showed improved prognostic accuracy.
Conclusions:
- The developed CTA tool offers a valuable method for stromal TIL assessment and prognosis in TNBC.
- Integrating VTA and CTA may enhance risk management and clinical decision-making for pathologists.
- This computational approach provides a promising advancement in TNBC biomarker analysis.
