Instance Segmentation of Multiple Myeloma Cells Using Deep-Wise Data Augmentation and Mask R-CNN.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Multiple myeloma diagnosis relies on manual microscopic analysis of bone marrow, which is time-consuming and prone to human error.
• Automated methods are needed to improve the efficiency and accuracy of multiple myeloma cell detection.

Purpose of the Study:

• To develop and evaluate a computer-aided detection and segmentation system for multiple myeloma cells.
• To enhance the performance of deep learning models for cell segmentation using novel data augmentation techniques.

Main Methods:

• Instance segmentation of multiple myeloma cells was performed using various Mask R-CNN models on original, contrast-enhanced, and stained microscopic images.
• A deep learning-based data augmentation method, termed deep-wise augmentation, was developed and applied to improve model performance.

Main Results:

• The Mask R-CNN model utilizing contrast-enhanced images combined with deep-wise augmentation achieved superior performance.
• This optimized model demonstrated a mean precision of 0.9973, mean recall of 0.8631, and mean intersection over union (IOU) of 0.9062.

Conclusions:

• The proposed computer-aided system, particularly the Mask R-CNN model with contrast enhancement and deep-wise augmentation, offers a highly effective solution for multiple myeloma cell detection.
• This approach has the potential to significantly improve diagnostic efficiency and accuracy in clinical practice.