Histopathology-based Protein Multiplex Generation using Deep Learning

Area of Science:

• Computational pathology
• Artificial intelligence in oncology
• Biomedical imaging analysis

Background:

• Multiplexed imaging is vital for tumor microenvironment (TME) analysis but faces adoption barriers.
• Current methods are limited by cost, time, and tissue requirements.
• Understanding TME interactions is crucial for cancer research and treatment.

Purpose of the Study:

• To introduce HistoPlexer, a deep learning framework for generating spatial protein multiplexes from histopathology images.
• To overcome the limitations of existing multiplexed imaging technologies.
• To provide a cost- and time-effective tool for TME analysis.

Main Methods:

• HistoPlexer utilizes conditional generative adversarial networks (cGANs) with custom loss functions.
• The framework mitigates slice-to-slice variations and preserves spatial protein correlations.
• Evaluation involved metastatic melanoma samples and an independent cohort.

Main Results:

• HistoPlexer outperformed existing methods in quantitative metrics (SSIM, PSNR).
• Expert evaluation confirmed high realism of generated protein multiplexes.
• Generated data enabled accurate tumor stratification (immune hot/cold) and improved survival prediction models.

Conclusions:

• HistoPlexer offers a cost- and time-effective solution for generating whole-slide protein multiplexes from H&E images.
• The framework accurately captures spatial protein co-localization and TME characteristics.
• HistoPlexer shows promise for advancing precision oncology by enhancing TME understanding.