Related Experiment Video
Updated: Nov 12, 2025
10:37
A Multimodal Imaging Framework to Advance Phenotyping of Living Label-free Breast Cancer Cells
Published on: August 22, 2025
756
A multi-phase deep CNN based mitosis detection framework for breast cancer histopathological images
Anabia Sohail1, Asifullah Khan2,3, Noorul Wahab4
1Pattern Recognition Lab, DCIS, Pakistan Institute of Engineering and Applied Sciences (PIEAS), Nilore, Islamabad, 45650, Pakistan.
Scientific Reports
|March 19, 2021
Summary
This study introduces MP-MitDet, a deep CNN framework for automated mitosis detection in breast cancer images. The system accurately identifies mitotic nuclei, improving tumor grading and reducing manual workload.
Area of Science:
- Computational pathology
- Artificial intelligence in oncology
- Image analysis for cancer diagnostics
Background:
- Accurate tumor grading relies on the mitotic activity index, a crucial prognostic factor.
- Manual detection of mitotic nuclei via microscopy is labor-intensive and requires automation.
- Histopathological image analysis presents challenges due to variations in mitotic nuclei appearance.
Purpose of the Study:
- To develop and evaluate a deep convolutional neural network (CNN) based framework, MP-MitDet, for automated mitosis detection in breast cancer histopathology.
- To improve the efficiency and accuracy of mitotic nuclei identification for prognostic assessment.
- To address the need for automated solutions in quantitative tumor grading.
Main Methods:
- A multi-phase framework (MP-MitDet) integrating label refinement, tissue-level region selection, blob analysis, and cell-level refinement.
- Development of an automatic label-refiner for training deep CNNs with weak labels.
- Utilizing a deep instance-based detection and segmentation model for region proposal.
- Employing a custom CNN classifier, MitosRes-CNN, for precise cell-level mitosis filtering.
Main Results:
- The MP-MitDet framework achieved strong performance on the TUPAC16 dataset, with an F-score of 0.75, recall of 0.76, precision of 0.71, and an AUC of 0.78.
- The MitosRes-CNN classifier demonstrated effective discrimination against false positives.
- The proposed framework showed promising generalization capabilities across heterogeneous mitotic nuclei.
Conclusions:
- The MP-MitDet framework offers an effective automated solution for mitotic nuclei identification in breast cancer histopathology.
- The developed methods, including the label-refiner and MitosRes-CNN, contribute to advancing automated cancer diagnostics.
- The results suggest the potential of this framework for improving prognostic accuracy and reducing pathologist workload.
