Deep Active Learning for Automatic Mitotic Cell Detection on HEp-2 Specimen Medical Images

Area of Science:

• Medical Imaging Analysis
• Computational Pathology
• Artificial Intelligence in Diagnostics

Background:

• Accurate identification of Human Epithelial Type 2 (HEp-2) mitotic cells is critical for anti-nuclear antibodies (ANAs) testing in diagnosing connective tissue diseases (CTD).
• Current manual screening methods for ANAs are labor-intensive, subjective, and have low throughput, necessitating automated solutions.
• Computer-aided diagnosis (CAD) systems are needed to improve the efficiency and reliability of HEp-2 cell analysis.

Purpose of the Study:

• To develop a deep active learning (DAL) framework for automated detection of mitotic cells in HEp-2 specimen images.
• To overcome the challenge of limited labeled cell data through an efficient active learning strategy.
• To directly identify mitotic cells in whole microscopic images, eliminating the need for a separate segmentation step.

Main Methods:

• A deep active learning (DAL) approach was implemented to train deep learning detectors for mitotic cell identification.
• The YOLO and Faster R-CNN deep learning models were adapted to directly detect mitotic cells in HEp-2 images.
• The framework was evaluated on the I3A Task-2 dataset using 5-fold cross-validation, with iterative data labeling rounds.

Main Results:

• The YOLO predictor achieved high performance with average scores of 90.011% recall, 88.307% precision, and 81.531% mAP.
• The Faster R-CNN predictor demonstrated strong results with average scores of 86.986% recall, 85.282% precision, and 78.506% mAP.
• The DAL method significantly improved prediction accuracy by enhancing data annotation quality over four labeling rounds.

Conclusions:

• The proposed DAL framework effectively automates the detection of HEp-2 mitotic cells, addressing the limitations of manual screening.
• The system demonstrates practical applicability for assisting medical professionals in rapid and accurate diagnostic decisions.
• This approach enhances the throughput and objectivity of ANAs testing, potentially improving the diagnosis of connective tissue diseases.