Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Cell Maps Representation For Lung Adenocarcinoma Growth Patterns Classification In Whole Slide Images

Arwa Al-Rubaian, Gozde N. Gunesli, Wajd A. Althakfi, Ayesha Azam, Nasir Rajpoot, Shan E Ahmed Raza

TL;DR

This work tackles the challenge of classifying LUAD tissue growth patterns at the tile level within whole slide images, a task confounded by tumor heterogeneity. It introduces a cell maps representation, generated from H&E WSIs using Hover-Net to produce three nucleus-type masks that are stacked into a compact input for a CNN (ResNet50) trained to predict six classes (five patterns plus normal). Under proper WSI-based validation, the proposed method outperforms state-of-the-art baselines and demonstrates robust generalization on unseen slides, addressing biases common in tile-based splits. The approach has potential to improve prognostic assessments and streamline pathology workflows by focusing on cellular composition rather than solely tissue morphology.

Abstract

Lung adenocarcinoma is a morphologically heterogeneous disease, characterized by five primary histologic growth patterns. The quantity of these patterns can be related to tumor behavior and has a significant impact on patient prognosis. In this work, we propose a novel machine learning pipeline capable of classifying tissue tiles into one of the five patterns or as non-tumor, with an Area Under the Receiver Operating Characteristic Curve (AUCROC) score of 0.97. Our model's strength lies in its comprehensive consideration of cellular spatial patterns, where it first generates cell maps from Hematoxylin and Eosin (H&E) whole slide images (WSIs), which are then fed into a convolutional neural network classification model. Exploiting these cell maps provides the model with robust generalizability to new data, achieving approximately 30% higher accuracy on unseen test-sets compared to current state of the art approaches. The insights derived from our model can be used to predict prognosis, enhancing patient outcomes.

Cell Maps Representation For Lung Adenocarcinoma Growth Patterns Classification In Whole Slide Images

TL;DR

This work tackles the challenge of classifying LUAD tissue growth patterns at the tile level within whole slide images, a task confounded by tumor heterogeneity. It introduces a cell maps representation, generated from H&E WSIs using Hover-Net to produce three nucleus-type masks that are stacked into a compact input for a CNN (ResNet50) trained to predict six classes (five patterns plus normal). Under proper WSI-based validation, the proposed method outperforms state-of-the-art baselines and demonstrates robust generalization on unseen slides, addressing biases common in tile-based splits. The approach has potential to improve prognostic assessments and streamline pathology workflows by focusing on cellular composition rather than solely tissue morphology.

Abstract

Lung adenocarcinoma is a morphologically heterogeneous disease, characterized by five primary histologic growth patterns. The quantity of these patterns can be related to tumor behavior and has a significant impact on patient prognosis. In this work, we propose a novel machine learning pipeline capable of classifying tissue tiles into one of the five patterns or as non-tumor, with an Area Under the Receiver Operating Characteristic Curve (AUCROC) score of 0.97. Our model's strength lies in its comprehensive consideration of cellular spatial patterns, where it first generates cell maps from Hematoxylin and Eosin (H&E) whole slide images (WSIs), which are then fed into a convolutional neural network classification model. Exploiting these cell maps provides the model with robust generalizability to new data, achieving approximately 30% higher accuracy on unseen test-sets compared to current state of the art approaches. The insights derived from our model can be used to predict prognosis, enhancing patient outcomes.
Paper Structure (11 sections, 2 figures, 1 table)

This paper contains 11 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. The Proposed Method
  3. Cell maps construction
  4. Growth pattern classification
  5. EXPERIMENTS
  6. Dataset
  7. Results
  8. Weak validation
  9. Strong validation
  10. CONCLUSIONs
  11. Acknowledgments

Figures (2)

  • Figure 1: Sample images of lung adenocarcinoma (LUAD) growth patterns. (a) shows the H&E image, (b) shows the corresponding cell map, where green represents neoplastic cells, red shows connective cells, and blue shows non-neoplastic cells
  • Figure 2: An overview of the proposed Model. First the WSI is divided into tiles, which are then processed using HoverNet to obtain cell locations and types, then a cell map is generated by stacking the masks of three cell types (neoplastic, non-neoplastic, connective). The generated cell map is used as an input to a Convolutional neural network (ResNet50) for growth pattern classification. * Colors of the cell masks are inverted for illustration purposes