A novel cost function for nuclei segmentation and classification in imbalanced histopathology data-sets.

Area of Science:

• Medical imaging
• Computational pathology
• Artificial intelligence in oncology

Background:

• Histopathological analysis is vital for cancer diagnosis, staging, and treatment planning.
• Detecting and classifying rare cells in imbalanced datasets poses a significant challenge in cancer diagnostics.
• Early cancer detection is crucial for reducing mortality and improving patient quality of life.

Purpose of the Study:

• To develop and evaluate SPNet, a spatially aware convolutional neural network, for improved detection and classification of rare cells in histopathological images.
• To address the challenge of imbalanced datasets in cancer pathology by employing a novel spatial data balancing technique.
• To enhance the accuracy of cancer diagnosis and streamline the diagnostic process.

Main Methods:

• Development of SPNet, a spatially aware convolutional neural network incorporating a novel cost function targeting spatial regions.
• Implementation of a spatial data balancing technique to address imbalanced datasets.
• Integration of SPNet with a ResNet50-SE encoder and evaluation on the CoNSeP dataset.

Main Results:

• SPNet enhanced the classification of rare nuclei by 21.8% through spatial data balancing.
• The novel cost function of SPNet resulted in a 1.9% increase in F1 classification for rare cell types within the CoNSeP dataset.
• Integration with ResNet50-SE improved the mean F1 score for classifying all nuclei by 4.3% compared to HoVer-Net.

Conclusions:

• SPNet effectively addresses the challenge of classifying rare cells in imbalanced histopathological datasets.
• The model demonstrates significant improvements in accuracy for both rare cell classification and overall nuclei classification.
• SPNet's integration into medical devices could streamline diagnostics and minimize false negatives, improving patient outcomes.