A Novel Encoding and Decoding Calibration Guiding Pathway for Pathological Image Analysis.

Area of Science:

• Digital pathology
• Computational imaging
• Artificial intelligence in medicine

Background:

• Diagnostic pathology relies on accurate image analysis for carcinoma identification.
• Encoder-decoder architectures (EDAs) using convolutional neural networks (CNNs) are prevalent in pathological image analysis.
• Existing EDAs struggle with complex backgrounds, hindering focus on regions of interest (RoIs) and impacting quantitative accuracy.

Purpose of the Study:

• To address the limitations of current EDAs in handling complex backgrounds in pathological images.
• To improve the accuracy and reliability of quantitative analysis in diagnostic pathology.
• To enhance the ability of CNN-based EDAs to focus on relevant regions of interest (RoIs).

Main Methods:

• Development and implementation of a novel pathway named Global Bank (GLB).
• Integration of the GLB pathway to guide the encoder and decoder components of CNN-based EDAs.
• Extensive experimental validation using various commonly used EDAs on pathological images.

Main Results:

• The GLB-enhanced architecture demonstrated improved performance in analyzing pathological images.
• The proposed method effectively guided feature extraction towards regions of interest (RoIs).
• Quantitative results obtained using the GLB-modified architecture were significantly more accurate compared to baseline EDAs.

Conclusions:

• The Global Bank (GLB) pathway offers a significant improvement for CNN-based encoder-decoder architectures in pathological image analysis.
• GLB enhances the focus on regions of interest (RoIs), overcoming challenges posed by complex backgrounds.
• This advancement leads to more reliable and accurate quantitative pathology, supporting improved diagnostic capabilities.