Quality Control in Digital Pathology: Automatic Fragment Detection and Counting.

Area of Science:

• Digital Pathology
• Computational Pathology
• Machine Learning in Histopathology

Background:

• Manual assessment of tissue fragments in pathology specimen processing is critical but time-consuming.
• Current manual methods can lead to delays in case evaluation and potential loss of valuable material.
• Automation is needed to improve efficiency and accuracy in fragment assessment.

Purpose of the Study:

• To develop an autonomous system for detecting and counting tissue fragments on pathology slides.
• To compare the performance of conventional machine learning methods against deep learning models for this task.
• To optimize fragment detection in colorectal biopsy and polypectomy specimens.

Main Methods:

• Developed an autonomous system for fragment detection and counting.
• Compared a two-stage conventional machine learning approach (supervised classifier + hierarchical clustering) with deep learning models (Fast R-CNN, YOLOv5).
• Trained and tested models on a dataset of 1276 manually labeled images of colorectal specimens.

Main Results:

• The YOLOv5 deep learning architecture achieved the highest performance.
• YOLOv5 demonstrated a mean average precision (mAP@0.5) of 0.977 for fragment/set detection.
• Deep learning methods significantly outperformed conventional machine learning approaches.

Conclusions:

• An autonomous system utilizing YOLOv5 can accurately detect and count tissue fragments in pathology slides.
• This automated approach offers a significant improvement over manual assessment, reducing processing time.
• The developed system has the potential to streamline digital pathology workflows and enhance diagnostic efficiency.