LIVER STEATOSIS SEGMENTATION WITH DEEP LEARNING METHODS.

Area of Science:

• Medical Imaging
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Liver steatosis, or fatty liver disease, involves abnormal lipid accumulation in liver cells.
• Accurate quantification of steatosis is crucial for diagnosing liver diseases and evaluating liver transplants.
• Precise segmentation of steatosis droplets in histopathological images is challenging due to droplet overlap.

Purpose of the Study:

• To develop and evaluate a deep learning model for segmenting steatosis droplets in liver microscopy images.
• To address the challenge of segmenting highly overlapped steatosis regions.
• To assess the model's performance in quantifying steatosis for clinical applications.

Main Methods:

• Utilized the Mask R-CNN deep learning model, an extension of Faster R-CNN, for object detection and instance segmentation.
• Employed transfer learning to enhance the model's segmentation capabilities.
• Applied the model to segment steatosis droplets within liver histopathological microscopy images.

Main Results:

• The Mask R-CNN model achieved an Average Precision of 75.87% for steatosis segmentation.
• Performance metrics included a Recall of 60.66%, an F1-score of 65.88%, and a Jaccard index of 76.97%.
• The model demonstrated effectiveness in segmenting overlapped steatosis regions.

Conclusions:

• The deep learning-based Mask R-CNN model shows significant promise for accurate steatosis quantification in liver images.
• This approach can support clinical decision-making in liver disease diagnosis and allograft rejection prediction.
• Automated segmentation of liver steatosis can improve efficiency and accuracy in histopathological analysis.