LOGSAGE: LOG-BASED SALIENCY FOR GUIDED ENCODING IN ROBUST NUCLEI SEGMENTATION OF IMMUNOFLUORESCENCE HISTOLOGY IMAGES.

Area of Science:

• Computational biology
• Medical imaging analysis
• Cancer research

Background:

• The tumor microenvironment (TME) is crucial for cancer progression and treatment response.
• Accurate segmentation of TME cellular architecture in immunofluorescence (IF) images is complex and challenging.

Purpose of the Study:

• To introduce LoGSAGE-Net, a novel deep learning model for enhanced immunofluorescence image segmentation.
• To improve the accuracy and robustness of tumor microenvironment profiling.

Main Methods:

• Developed LoGSAGE-Net, coupling a Swin Transformer with Laplacian of Gaussian (LoG) saliency.
• Incorporated Dice and curvature alignment loss into the model's loss function.
• Applied the model to a large cohort of preclinical immunofluorescence data.

Main Results:

• LoGSAGE-Net demonstrated superior performance compared to state-of-the-art methods.
• Achieved a high Dice score of 94.92% for image segmentation.
• Obtained a Panoptic Quality (PQ) score of 81%, indicating excellent segmentation and classification.

Conclusions:

• LoGSAGE-Net provides a robust solution for segmenting complex TME images.
• The model facilitates accurate TME profiling, supporting the development of sensitive diagnostic and therapeutic assays.
• This advancement aids in understanding cancer progression and treatment response.