MIHIC: a multiplex IHC histopathological image classification dataset for lung cancer immune microenvironment...

Area of Science:

• Oncology
• Computational Pathology
• Bioinformatics

Background:

• Immunohistochemistry (IHC) is crucial for cancer diagnosis, enabling visualization of protein targets in tissue samples.
• Deep learning models can quantify the tumor immune microenvironment (TIME) in digitized IHC slides.
• A lack of publicly available IHC datasets hinders in-depth TIME analysis.

Purpose of the Study:

• To introduce the Multiplex IHC Histopathological Image Classification (MIHIC) dataset for TIME analysis in lung cancer.
• To benchmark deep learning models (CNNs and transformers) for classifying IHC stained histological images.
• To quantify TIME variables and assess their prognostic value for patient survival.

Main Methods:

• Creation of the MIHIC dataset with 309,698 multiplex IHC image patches, manually annotated by pathologists across seven tissue types.
• Benchmarking convolutional neural networks (CNNs) and transformer models for image classification tasks.
• Quantification of TIME variables using the best-performing model and correlation with patient survival outcomes.

Main Results:

• Transformer models showed slightly superior performance compared to CNNs in histological image classification, achieving a maximum accuracy of 0.811.
• Quantified TIME variables, specifically immune cells over stroma and tumor over tissue core, demonstrated prognostic value.
• The MIHIC dataset is the first publicly available lung cancer IHC dataset with 12 stains and multi-pathologist annotations.

Conclusions:

• The MIHIC dataset facilitates research into novel TIME quantification techniques and understanding immune-tumor interactions.
• Deep learning models, particularly transformers, show promise in analyzing complex IHC data for lung cancer.
• Automated TIME quantification from IHC images can predict patient survival, offering valuable clinical insights.