Cross-organ Deployment of EOS Detection AI without Retraining: Feasibility and Limitation

TL;DR

To determine the extent to which Eos segmentation models developed for the GI tract can be adapted to nasal applications without retraining, a CircleSnake model initially trained on upper-GI data is leveraged to segment Eos cells in whole slide images of nasal tissues.

Abstract

Chronic rhinosinusitis (CRS) is characterized by persistent inflammation in the paranasal sinuses, leading to typical symptoms of nasal congestion, facial pressure, olfactory dysfunction, and discolored nasal drainage, which can significantly impact quality-of-life. Eosinophils (Eos), a crucial component in the mucosal immune response, have been linked to disease severity in CRS. The diagnosis of eosinophilic CRS typically uses a threshold of 10-20 eos per high-power field (HPF). However, manually counting Eos in histological samples is laborious and time-intensive, making the use of AI-driven methods for automated evaluations highly desirable. Interestingly, eosinophils are predominantly located in the gastrointestinal (GI) tract, which has prompted the release of numerous deep learning models trained on GI data. This study leverages a CircleSnake model initially trained on upper-GI data to segment Eos cells in whole slide images (WSIs) of nasal tissues. It aims to determine the extent to which Eos segmentation models developed for the GI tract can be adapted to nasal applications without retraining. The experimental results show promising accuracy in some WSIs, although, unsurprisingly, the performance varies across cases. This paper details these performance outcomes, delves into the reasons for such variations, and aims to provide insights that could guide future development of deep learning models for eosinophilic CRS.

Figures (5)

• Figure 1: The goal of this study is to determine whether and under what conditions the pre-trained CircleSnake model for eosinophil detection can be directly applied to other parts of the body, such as from the esophagus to the nose.
• Figure 2: A pipeline of the CircleSnake segmentation model's network structure. The initial boundary forms a circular contour after verification by CircleSnake. Through a graph convolutional network (GCN), the outline eventually transforms into the final circular boundary of the eosinophil cell.
• Figure 3: The bar chart displays a comparison between counts obtained by human experts and those generated by the CircleSnake model. Each bar represents the average number of eosinophilic cells in the top five regions with the highest cell counts within a single high-power field (HPF) across all tissue sections for each patient. The CircleSnake model's data is shown in blue bars, while the data from the pathologists is depicted in orange bars. The data is arranged in ascending order based on the counts from the pathologists, while the threshold is set to 0.15.
• Figure 4: This scatter plot illustrates the comparison between human and machine counts. The blue line represents the linear regression, and the blue points indicate the coordinates of the counts, while the shaded region around the regression line depicts the confidence interval.
• Figure 5: This figure shows the visualization of all WSIs in the same location as Fig. \ref{['fig:LRL']}. The observations and interpretation are presented in the discussion section.