Using the low-pass monogenic signal framework for cell/background classification on multiple cell lines in...

Area of Science:

• Microscopy
• Image Analysis
• Computational Biology

Background:

• Bright-field microscopy often uses image defocusing to enhance cell contrast.
• The Transport of Intensity Equation (TIE) relates image contrast to physical light phase.
• The monogenic signal framework offers an approximation for solving the TIE.

Purpose of the Study:

• To evaluate if local phase derived from the monogenic signal framework improves cell/background classification accuracy.
• To compare the performance of local phase against defocused images in cell detection.

Main Methods:

• Tested on three cell lines (CHO, L929, Sf21) using bright-field microscopy images.
• Generated local phase and energy images via the low-pass monogenic signal framework.
• Employed machine learning classifiers (Random Forest, SVM with linear and RBF kernels) for classification.

Main Results:

• Local phase significantly improved classification accuracy over defocused images (7.3% RF, 11.6% linear SVM, 10.2% RBF SVM).
• Discriminative power order: at-focus signal < local energy < defocused signal < local phase.
• Local energy showed superiority over defocused signal for suspended cells.

Conclusions:

• Local phase computed via the low-pass monogenic signal framework is superior to defocused images for cell/background classification.
• This approach enhances accuracy in pixel-patch classification for bright-field microscopy.