Deep Learning-Based Microscopic Cell Detection Using Inverse Distance Transform and Auxiliary Counting.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Machine Learning

Background:

• Microscopic cell detection is hindered by occlusions and varied cell shapes in dense populations.
• Accurate cell counting and localization are critical for biomedical research and diagnostics.

Purpose of the Study:

• To develop an advanced automated cell detection framework.
• To improve accuracy and reduce false positives in microscopic cell analysis.

Main Methods:

• A deep learning model generating an inverse distance transform (IDT) map for cell instance highlighting.
• A secondary network regressing a cell density map for accurate cell counting.
• A counting-aided cell center extraction strategy refining detection using density map integration.

Main Results:

• Achieved high F-scores: 96.93% (VGG), 91.21% (MBM), and 92.00% (ADI), outperforming state-of-the-art methods.
• Demonstrated the lowest distance error, confirming precise cell localization.
• Significantly reduced false detection responses, enhancing overall accuracy.

Conclusions:

• The proposed framework offers a robust solution for automated cell detection in challenging microscopic images.
• This approach shows significant potential for advancing automated cell analysis in various biomedical applications.
• The integration of IDT and density maps provides a powerful tool for accurate cell counting and instance segmentation.