Singapore General Hospital (SGH)

Identified Problem

Distinguishing between fibroadenomas (FAS) and phyllodes tumors (PTs) in breast tissue samples is challenging due to overlapping features, yet they require different clinical management. Pathologists need to examine tissue samples under a microscope to make the distinction, but this process can be time-consuming and prone to human error. There is a need for a more efficient and accurate method of analyzing these samples.

AI Solution

A computer vision model was developed to analyze whole slide images of tissue samples to distinguish between FAS and PTs. This AI solution acts as a “second eye” for pathologists, working alongside them to improve diagnostic accuracy. The model works in two stages: first, it generates smaller image patches from the large whole slide images to identify the characteristics of the two types of regions, and second, it combines these analyses to predict whether the entire slide is FAS or PT.

Key Features

• The AI model uses computer vision to analyze digitized whole slide images.
• It generates smaller image patches from large whole slide images to identify the characteristics of FAS and PT regions.
• The model combines the analysis of these patches to make predictions about the entire slide.
• The AI model provides results that pathologists can incorporate into their own assessments.

Benefits of AI Solution

• The AI model helps improve diagnostic accuracy and reduce human errors.
• The AI model is faster, taking about five minutes for model inference, compared to 15 minutes for initial evaluation by pathologists, leading to potential cost savings.
• The model can help reduce the need for surgical management and reduce patient anxiety.
• The AI model achieved an overall diagnostic accuracy of 87.5%, with 80% and 95% accuracy in distinguishing FAS and PTs, respectively.

Impact of AI Solution

• The AI solution can be readily deployed not only in the SGH laboratory but also in other laboratories in Singapore and the region.
• The project highlights the importance of collaboration between pathologists and AI experts, combining their expertise to meet project objectives.
• The AI model was developed within a relatively short period of time with high quality of output.
• The project received recognition through publication in the Laboratory Investigation Journal, a reputable journal in the field.