Scalable histopathological image analysis via supervised hashing with multiple features.

Area of Science:

• Digital pathology
• Computational biology
• Medical image analysis

Background:

• Histopathology is vital for cancer diagnosis but interpretation is challenging.
• Digital histopathology and hashing-based retrieval offer efficiency and scalability.
• Feature fusion is underutilized in hashing for histopathological image analysis.

Purpose of the Study:

• To develop a hashing framework that integrates complementary features for improved histopathological image retrieval.
• To address the semantic gap between low-level features and high-level cancer diagnosis.
• To enhance the efficiency and accuracy of content-based image retrieval in digital pathology.

Main Methods:

• Exploited joint kernel-based supervised hashing (JKSH) to integrate features.
• Designed hashing functions using linearly combined kernel functions for individual features.
• Incorporated supervised information and utilized alternating optimization for learning.

Main Results:

• JKSH compresses high-dimensional features into binary bits for fast retrieval.
• Achieved 88.1% retrieval precision and 91.3% classification accuracy on 3121 breast-tissue images.
• Demonstrated query times of 16.5 ms, outperforming traditional methods.

Conclusions:

• JKSH effectively integrates complementary features for enhanced histopathological image analysis.
• The method significantly improves retrieval performance and classification accuracy in digital pathology.
• This approach offers a scalable and efficient solution for computer-aided cancer diagnosis.