Automatic enhancement preprocessing for segmentation of low quality cell images.

Area of Science:

• Biomedical Imaging
• Computational Biology
• Machine Learning

Background:

• Microscopic cell imaging under intense light poses challenges, often resulting in low-quality images.
• Low image quality significantly reduces the accuracy of semantic segmentation for cell analysis.
• Classical image preprocessing techniques are insufficient for effectively handling these low-quality cell images.

Purpose of the Study:

• To develop a novel automatic preprocessing technique for enhancing low-quality cell images.
• To introduce an automatic weighted ensemble learning method for improving segmentation accuracy.
• To address the limitations of existing methods in segmenting challenging microscopic cell images.

Main Methods:

• Proposed automatic enhancement preprocessing (AEP) using two deep neural networks to translate low-quality images into deep learning-friendly formats.
• Utilized penultimate feature maps from the first network as filters for image translation.
• Developed automatic weighted ensemble learning (AWEL) to aggregate multiple segmentation results by determining optimal weights.

Main Results:

• AEP successfully translated low-quality cell images into formats easily recognized and segmented by deep learning models.
• Experiments demonstrated a significant improvement in segmentation accuracy when using AEP in conjunction with AWEL.
• The proposed methods proved effective on two distinct cell image segmentation tasks.

Conclusions:

• The novel AEP technique effectively enhances low-quality cell images for improved deep learning-based segmentation.
• AWEL provides a robust approach to combine multiple segmentation outputs, further boosting accuracy.
• This combined approach offers a promising solution for accurate cell segmentation from challenging microscopic imagery.