Analysis and Validation of Image Search Engines in Histopathology

Area of Science:

• Digital pathology
• Computational biology
• Medical imaging analysis

Background:

• Accurate comparison of patient tissues using whole slide images (WSIs) is vital for diagnosis and prognosis.
• Existing WSI search methods require thorough evaluation for clinical applicability.

Purpose of the Study:

• To extensively analyze and validate four WSI search methods: bag of visual words (BoVW), Yottixel, SISH, and RetCCL.
• To assess the performance, accuracy, and efficiency of these search algorithms on diverse histopathology datasets.

Main Methods:

• Evaluation of four WSI search algorithms (BoVW, Yottixel, SISH, RetCCL) and their variants.
• Utilized four internal and three public datasets comprising over 200,000 patches from 38 classes/subtypes across five primary sites.
• Assessed algorithms based on accuracy, efficiency, speed, and consistency.

Main Results:

• BoVW demonstrated high efficiency and speed but low accuracy.
• Yottixel provided a balance of efficiency, speed, and moderate accuracy.
• SISH showed inefficiency and inconsistent results, while RetCCL was inadequate in both accuracy and efficiency.

Conclusions:

• No single evaluated method perfectly balances accuracy and efficiency for histopathology image search.
• Yottixel presents a viable option for moderately accurate and efficient WSI searching.
• Further research is needed to develop methods that optimize both accuracy and storage requirements for large-scale histopathological image analysis.