Automatic Quantification of Tumour Hypoxia From Multi-Modal Microscopy Images Using Weakly-Supervised Learning...

Area of Science:

• Oncology
• Computational Pathology
• Medical Imaging

Background:

• Hypoxia-modified cancer therapies show improved patient outcomes.
• Accurate measurement of tumor hypoxia is crucial for developing these therapies.
• Standardized methods for quantifying tumor hypoxia are lacking in clinical practice.

Purpose of the Study:

• To develop and validate machine learning methodologies for automated quantification of hypoxic regions in tumor histological images.
• To address the challenge of weakly-supervised learning due to limited annotation data.
• To compare the performance of four proposed machine learning approaches.

Main Methods:

• Four machine learning methodologies were proposed: naive grid classifier, baseline structured output learning, latent structured output learning with flexible graphs, and fully-convolutional neural networks for pixel-wise labeling.
• Training utilized a weakly-supervised approach with a high-order loss function based on manual annotations of hypoxic region counts.
• The methods were evaluated on 89 weakly annotated image pairs from eight tumors.

Main Results:

• Latent structured output learning (method 3) and fully-convolutional neural networks (method 4) demonstrated competitive and superior performance compared to naive (1) and baseline (2) methods.
• All proposed methodologies exhibited high correlation with clinical annotations.
• The developed methods effectively automate the quantification of normoxic, chronically hypoxic, and acutely hypoxic regions.

Conclusions:

• Advanced machine learning, particularly latent structured output learning and fully-convolutional networks, can effectively automate the quantification of tumor hypoxia from histological images.
• These automated methods show high accuracy and correlation with manual annotations, offering a standardized approach for clinical research.
• The findings support the development and validation of hypoxia-modified cancer therapies by providing reliable tumor hypoxia measurements.