Efficient mitosis detection: leveraging pre-trained faster R-CNN and cell-level classification.

Area of Science:

• Pathology
• Computer Vision
• Machine Learning

Background:

• Accurate assessment of mitotic activity is crucial for breast cancer grading and treatment.
• Manual counting of mitoses in Hematoxylin and Eosin (H&E)-stained slides is time-consuming and challenging.
• Computer-assisted methods can simplify mitosis detection.

Purpose of the Study:

• To investigate automatic mitosis detection in histopathology images using deep neural networks.
• To treat mitosis detection as an object detection problem.
• To evaluate the performance of pre-trained Faster R-CNN with raw image data for mitosis detection.

Main Methods:

• Utilized multiple neural networks for mitosis detection.
• Employed Faster R-CNN for object detection on histopathology images.
• Conducted experiments on the MITOS-ATYPIA-14 and TUPAC16 datasets.

Main Results:

• Demonstrated the potential of deep neural networks for automatic feature extraction.
• Achieved comparable or improved results against existing methods in literature.
• Validated the effectiveness of Faster R-CNN for mitosis detection in tissue samples.

Conclusions:

• Deep learning approaches show promise for automated mitosis detection in breast cancer pathology.
• Object detection frameworks like Faster R-CNN are effective tools for this task.
• Automated mitosis counting can enhance the efficiency and accuracy of cancer grading.