Detection on Cell Cancer Using the Deep Transfer Learning and Histogram Based Image Focus Quality Assessment

Area of Science:

• Digital pathology
• Computational imaging
• Artificial intelligence in medicine

Background:

• Whole-slide imaging (WSI) is increasingly used for AI development in histopathology.
• Automated quality control is crucial for clinical implementation of digital pathology due to scanner inaccuracies and large file sizes.
• Image blur from inaccurate focusing can render digitized slides unusable for AI analysis.

Purpose of the Study:

• To develop and validate a computational metric for assessing digital pathology image focus quality.
• To enable automated quality control for whole-slide images (WSIs) used in artificial intelligence (AI) development.
• To improve the reliability and clinical relevance of digital pathology data.

Main Methods:

• Proposed a novel metric using even-derivative filter bases to create a human visual system-like kernel.
• Applied the kernel to assess patch-level focus quality by analyzing high-frequency image data degraded by scanner optics.
• Developed a heatmap for local slide-level focus quality assessment.

Main Results:

• The proposed focus quality metric demonstrated better correlation with ground-truth z-level data than previous methods.
• The system achieved high accuracy in identifying cancer cells using deep learning techniques (e.g., GoogleNet at 98.5%).
• The focus quality heatmap proved valuable for automated slide quality control, aligning with subjective ratings.

Conclusions:

• The developed computational method provides an efficient and reliable approach for assessing digital pathology image focus quality.
• Automated quality control using this technique can overcome a significant barrier to the clinical implementation of digital pathology.
• This approach enhances the usability of digitized histopathology slides for AI-driven diagnostic tools.