Enhanced Nuclei Segmentation in Histopathology Image Leveraging RGB Channels through Triple-Encoder and Single-Decoder Architectures

In biomedical imaging, nucleus segmentation is critical for understanding cellular structure and function. This work proposes a novel approach to nucleus segmentation that exploits an image’s red, green, and blue channels as input. We use three Encoders for each channel and a single Decoder to reconstruct the original input image. Associating an Encoder with each of the three-color channels in our design allows us to combine them in a particular way that fully utilizes the RGB color space’s information. Flowing up-sampling for each color channel image decodes and reconstructs the original idea through the skip connection. Our model was successfully trained and evaluated using numerous datasets from Kaggle. To perform the segmentation task, we used a small publicly accessible prostate cancer dataset with more than 17000 labeled nuclei. Our model performed remarkably well on both datasets, obtaining a 99.98% accuracy, a 97% Dice coefficient, and a 95% Jaccard coefficient. These outstanding measures demonstrate the model’s capacity to provide cutting-edge nucleus segmentation. Our method gives fresh perspectives and establishes a standard for nucleus segmentation tasks by exploiting the rich information within the RGB channels. The success of our approach implies that incorporating data from many channels can produce more precise and dependable segmentation, creating new opportunities for biomedical image analysis.

Перевод пока недоступен