Time-efficient sparse analysis of histopathological whole slide images

Area of Science:

• Digital pathology
• Computer vision in medicine
• Medical image analysis

Background:

• Histopathological examination of whole slide images (WSI) is crucial for disease prognosis but remains labor-intensive for human experts.
• Advances in scanning technology have not fully automated WSI analysis, highlighting a need for computational solutions.

Purpose of the Study:

• To develop an innovative platform for fast, multi-scale analysis of histopathological WSI.
• To reduce the time required for WSI analysis to match pathologist performance.

Main Methods:

• Implementation of a multi-scale computer vision framework utilizing application-driven (high-resolution) and generic (low-resolution) algorithms.
• Integration of sparse coding and dynamic sampling techniques for efficient image analysis.
• Leveraging Graphics Processing Units (GPUs) to enhance system time-efficiency.

Main Results:

• The platform successfully identifies high-power fields of interest for global grading in histopathology.
• Automatic WSI analysis time was comparable to that of human pathologists, averaging around ten minutes per WSI.
• The system was validated in a computer-aided breast biopsy analysis application.

Conclusions:

• The proposed multi-scale computer vision platform significantly accelerates WSI analysis.
• This methodology offers a major contribution towards efficient and automated histopathological assessment.
• The system demonstrates potential to aid pathologists in disease prognosis and grading.