Segmentation and counting of multiple myeloma cells using IEMD based deep neural network

Area of Science:

• Biomedical image analysis
• Computational pathology
• Deep learning in healthcare

Background:

• Accurate segmentation of cell nuclei in microscopic images is crucial for early disease prediction.
• Multiple myeloma, a plasma cell cancer, requires precise cell and nucleus segmentation for detection.
• Existing methods face challenges in reliably identifying and quantifying myeloma cells.

Purpose of the Study:

• To develop an advanced deep learning framework for segmenting and counting multiple myeloma cells.
• To improve the accuracy and efficiency of multiple myeloma detection using biomedical image analysis.
• To aid in the early prediction and diagnosis of multiple myeloma.

Main Methods:

• Designed two modules: one for nucleus recognition using a deep Intrinsic Mode Function Decomposition (IMFD) neural network, and another for cytoplasm differentiation.
• Utilized different Intrinsic Mode Functions (IMFs) for detailed frequency component extraction and feature enhancement.
• Developed a novel counting algorithm for myeloma-affected plasma cells using the Python TensorFlow framework.

Main Results:

• The proposed deep learning approach demonstrated superior performance in myeloma recognition and detection compared to existing methods.
• Experimental results on the SegPC dataset validated the effectiveness of the image segmentation mechanism.
• The method successfully segmented nuclei and cytoplasm, enabling accurate cell counting.

Conclusions:

• The developed image segmentation and counting mechanism offers a significant advancement in multiple myeloma detection.
• Early identification of multiple myeloma through this advanced technique can improve patient prognosis and treatment success.
• This research provides a robust tool for biological researchers in computer-assisted healthcare systems.