Development and Validation an Integrated Deep Learning Model to Assist Eosinophilic Chronic Rhinosinusitis Diagnosis:...

Area of Science:

• Otorhinolaryngology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate non-invasive preoperative prediction of eosinophilic chronic rhinosinusitis (eCRS) is lacking.
• Current diagnosis relies on invasive histopathological analysis.
• Need for advanced diagnostic tools to guide personalized therapy.

Purpose of the Study:

• To develop an integrated deep learning model for non-invasive preoperative identification of eCRS using CT images and clinical data.
• To explore the biological underpinnings of eCRS prediction through proteomic analysis.

Main Methods:

• Utilized CT images from 1098 patients across two hospitals.
• Employed deep learning models (3D-ResNet, 3D-Xception, HR-Net) for feature extraction.
• Integrated deep learning scores with clinical data using a support vector machine for classification.
• Performed proteomic analysis on a subset of patients to identify dysregulated genes and pathways.

Main Results:

• The integrated deep learning model achieved an Area Under the Curve (AUC) of 0.851 (internal) and 0.821 (external) for eCRS prediction.
• Proteomic analysis identified 594 dysregulated genes in predicted eCRS patients, linked to pathways like chemokine signaling.

Conclusions:

• The developed integrated deep learning model offers an effective, non-invasive method for preoperative eCRS prediction.
• This approach facilitates personalized therapeutic strategies and contributes to precision medicine in CRS management.