Related Experiment Video
Updated: May 9, 2025
08:54
Pathological Analysis of Lung Metastasis Following Lateral Tail-Vein Injection of Tumor Cells
Published on: May 20, 2020
8.7K
CellOMaps: A compact representation for robust classification of lung adenocarcinoma growth patterns
Arwa Al-Rubaian1, Gozde N Gunesli1, Wajd A Althakfi2
1Tissue Image Analytics Centre, Department of Computer Science, University of Warwick, UK.
Computers in Biology and Medicine
|May 1, 2025
Summary
This study introduces a new machine learning pipeline for classifying lung adenocarcinoma (LUAD) growth patterns from whole slide images. The method achieves state-of-the-art performance and shows potential for predicting tumor mutational burden.
Area of Science:
- Digital Pathology
- Computational Oncology
- Machine Learning in Medicine
Background:
- Lung adenocarcinoma (LUAD) exhibits significant morphological heterogeneity with five primary histological growth patterns.
- Accurate classification of LUAD growth patterns is vital for prognosis but is challenged by subjectivity and observer variability.
- Existing machine learning approaches often focus on predominant pattern classification at the slide level.
Purpose of the Study:
- To develop a generalizable machine learning pipeline for classifying LUAD histological growth patterns.
- To address the limitations of subjectivity and observer variability in LUAD pattern classification.
- To explore the utility of the pipeline's outputs for predicting tumor mutational burden (TMB).
Main Methods:
- Development of a novel compact Cell Organization Maps (cellOMaps) representation to capture spatial patterns from Hematoxylin and Eosin (H&E) whole slide images (WSIs).
- Implementation of a machine learning pipeline utilizing cellOMaps for classifying tissue into five LUAD patterns or as non-tumor.
- Evaluation of the pipeline's performance on internal unseen slides and external datasets.
Main Results:
- The proposed pipeline achieved state-of-the-art performance in LUAD growth pattern classification.
- The method demonstrated superior performance compared to current approaches on both internal and external validation datasets.
- Preliminary findings indicate the model's capability to predict patient Tumor Mutational Burden (TMB) levels.
Conclusions:
- The developed machine learning pipeline offers a robust and generalizable solution for objective LUAD histological growth pattern classification.
- The cellOMaps representation effectively captures crucial spatial information for accurate classification.
- This approach has potential implications for improving prognostic accuracy and guiding personalized treatment strategies through TMB prediction.
