MRI-based radiomics-deep learning model for preoperative pathogen prediction in perianal abscesses

Area of Science:

• Radiology
• Infectious Diseases
• Medical Imaging

Background:

• Perianal abscesses are often caused by Escherichia coli, necessitating accurate preoperative pathogen identification.
• Current diagnostic methods can be invasive and time-consuming, delaying appropriate treatment.

Purpose of the Study:

• To develop and validate a hybrid model integrating MRI-based radiomics, deep learning, and clinical variables.
• To achieve precise preoperative differentiation of Escherichia coli from other pathogens in perianal abscesses.

Main Methods:

• Retrospective analysis of 215 patients with culture-confirmed perianal abscesses.
• Extraction of radiomic and deep learning features from multi-sequence MRI (T1WI, T2WI, FS-T2WI).
• Development of a hybrid logistic regression model incorporating an MRI signature and clinical predictors, evaluated using ROC analysis and DCA.

Main Results:

• The hybrid model achieved an AUC of 0.885 in the testing set, outperforming the MRI signature alone (AUC=0.860).
• The model demonstrated high accuracy (0.815), sensitivity (0.818), and specificity (0.812).
• Gender and diabetes were identified as significant independent predictors.

Conclusions:

• The developed hybrid model effectively differentiates Escherichia coli in perianal abscesses using preoperative multi-sequence MRI.
• This noninvasive approach has significant clinical potential to guide antibiotic selection, moving towards precision medicine strategies.