[Paper Review] Bone marrow cells detection: A technique for the microscopic image analysis

Haichao Cao, Hong Liu+1 more|arXiv (Cornell University)|May 5, 2018

Digital Imaging for Blood Diseases3 references3 citations

TL;DR

This paper proposes a novel automated method for detecting and classifying bone marrow cells (BMCs) in microscopic images using a three-stage pipeline: localization, segmentation, and classification. The approach combines color transformation, stepwise averaging, Otsu's thresholding, region growing, K-means clustering, and marker-controlled watershed for segmentation, followed by SVM, Random Forest, and other classifiers; it achieves a highest average recognition rate of 87.49% using SVM.

ABSTRACT

In the detection of myeloproliferative, the number of cells in each type of bone marrow cells (BMC) is an important parameter for the evaluation. In this study, we propose a new counting method, which also consists of three modules including localization, segmentation and classification. The localization of BMC is achieved from a color transformation enhanced BMC sample image and stepwise averaging method (SAM). In the nucleus segmentation, both SAM and Otsu's method will be applied to obtain a weighted threshold for segmenting the patch into nucleus and non-nucleus. In the cytoplasm segmentation, a color weakening transformation, an improved region growing method and the K-Means algorithm are used. The connected cells with BMC will be separated by the marker-controlled watershed algorithm. The features will be extracted for the classification after the segmentation. In this study, the BMC are classified using the SVM, Random Forest, Artificial Neural Networks, Adaboost and Bayesian Networks into five classes including one outlier, namely, neutrophilic split granulocyte, neutrophilic stab granulocyte, metarubricyte, mature lymphocytes and the outlier (all other cells not listed). Our experimental results show that the best average recognition rate is 87.49% for the SVM.

Motivation & Objective

• To develop an automated system for accurate detection and classification of bone marrow cells in histopathological images.
• To address the challenge of manual counting in myeloproliferative disorder diagnosis, which is time-consuming and subjective.
• To improve segmentation accuracy by integrating multiple image processing techniques for nuclei and cytoplasm.
• To evaluate and compare multiple machine learning classifiers for BMC classification into five types plus an outlier class.
• To achieve a robust, repeatable method suitable for clinical diagnostics using standard microscopic imaging.

Proposed method

• Localization is performed using color transformation and the stepwise averaging method (SAM) on enhanced BMC images.
• Nuclear segmentation employs a weighted threshold derived from both SAM and Otsu’s method to distinguish nucleus from non-nucleus regions.
• Cytoplasm segmentation uses color weakening transformation, an improved region growing algorithm, and K-means clustering.
• Connected cells are separated using the marker-controlled watershed algorithm to resolve overlapping cells.
• Feature extraction is performed post-segmentation to feed into the classification stage.
• Five classifiers—SVM, Random Forest, Artificial Neural Networks, Adaboost, and Bayesian Networks—are used to classify BMCs into five types: neutrophilic split granulocyte, neutrophilic stab granulocyte, metarubricyte, mature lymphocytes, and an outlier class.

Experimental results

Research questions

• RQ1 Can a hybrid image processing pipeline improve the accuracy of bone marrow cell detection in microscopic images compared to conventional methods?
• RQ2 How effective is the combination of stepwise averaging and Otsu’s method in nucleus segmentation under varying image conditions?
• RQ3 To what extent does the improved region growing and K-means approach enhance cytoplasm segmentation accuracy?
• RQ4 Which machine learning classifier yields the highest recognition rate for bone marrow cell classification in this multi-class setting?
• RQ5 Can the marker-controlled watershed algorithm effectively separate overlapping bone marrow cells in dense microscopic fields?

Key findings

• The proposed method achieves a highest average recognition rate of 87.49% using the Support Vector Machine (SVM) classifier.
• The integration of color transformation, SAM, and Otsu’s method significantly improves nucleus segmentation accuracy.
• The improved region growing and K-means clustering techniques effectively segment cytoplasmic regions with minimal noise and over-segmentation.
• The marker-controlled watershed algorithm successfully resolves overlapping cells, reducing false positives in cell counting.
• Among the tested classifiers, SVM outperforms Random Forest, Artificial Neural Networks, Adaboost, and Bayesian Networks in classification accuracy.
• The overall pipeline demonstrates robustness and reproducibility on a dataset of bone marrow microscopic images, suitable for clinical deployment.