Semi-HIC: A novel semi-supervised deep learning method for histopathological image classification.

Area of Science:

• Digital pathology
• Machine learning in oncology
• Computational biology

Background:

• Histopathological images are crucial for cancer diagnosis, but supervised learning methods require extensive labeled data, which is costly and time-consuming to obtain.
• Semi-supervised learning offers a promising alternative to overcome data annotation limitations in histopathological image classification.
• Existing semi-supervised methods like Learning by Association (LA) face challenges with histopathological images due to high inter-class similarity and intra-class heterogeneity.

Purpose of the Study:

• To propose a novel semi-supervised deep learning method, Semi-HIC, for improved histopathological image classification.
• To address the limitations of existing methods in handling the complexities of histopathological data.
• To leverage large amounts of unlabeled data for enhanced cancer diagnosis.

Main Methods:

• Developed a new semi-supervised deep learning framework named Semi-HIC.
• Introduced a novel semi-supervised loss function combining association cycle consistency (ACC) and maximal conditional association (MCA) losses.
• Employed an efficient network architecture with cascaded Inception blocks (CIBs) for learning discriminative embeddings.

Main Results:

• The proposed Semi-HIC method effectively utilizes unlabeled histopathological image patches.
• The ACC and MCA losses successfully mitigate inter-class similarity and intra-class variation issues.
• Semi-HIC demonstrated superior performance compared to existing deep learning methods and the semi-supervised LA method on benchmark datasets (Bioimaging 2015 and BACH).

Conclusions:

• Semi-HIC offers a significant advancement in semi-supervised learning for histopathological image classification.
• The method's ability to handle data complexities and leverage unlabeled data leads to improved diagnostic accuracy.
• Semi-HIC provides a robust and efficient solution for cancer diagnosis using digital pathology, outperforming prior approaches.