Deep Learning-Inferred Multiplex ImmunoFluorescence for Immunohistochemical Image Quantification.

Area of Science:

• Computational pathology
• Biomedical image analysis
• Artificial intelligence in diagnostics

Background:

• Immunohistochemistry (IHC) is crucial for patient care but often reported qualitatively or semi-quantitatively.
• Current IHC reporting lacks precise, quantitative single-cell protein expression data.
• Multiplex immunofluorescence (mpIF) offers more informative data but is costly and less accessible.

Purpose of the Study:

• To develop a single-step deep learning solution for quantitative single-cell IHC scoring.
• To translate standard IHC images into informative mpIF-like data.
• To improve cell segmentation and protein quantification accuracy in pathology.

Main Methods:

• A multitask deep learning framework, DeepLIIF, was developed.
• A novel dataset of co-registered IHC and mpIF slides was created for training.
• A new nuclear-envelope stain, LAP2beta, was introduced to enhance cell segmentation.

Main Results:

• DeepLIIF successfully performs stain deconvolution, cell segmentation, and quantitative single-cell IHC scoring.
• The framework translates IHC to mpIF, providing ground truth for IHC channels.
• The model generalizes to various markers (Ki67, CD3, CD8, etc.) and noisy images, outperforming manual scoring.

Conclusions:

• DeepLIIF offers a powerful, automated approach for quantitative analysis of IHC slides.
• This method enhances diagnostic accuracy by providing precise single-cell protein quantification.
• The framework democratizes access to high-quality biomarker data from routine pathology.