Truace
Subscribe

CT-based deep learning for identifying sinonasal inverted papilloma: a single-center retrospective study

This study aimed to develop a CT-based deep learning model as a decision-support system for the classification of sinonasal inverted papilloma (IP), nasal polyps, and normal paranasal anatomy. A retrospective dataset of 871 patients who underwent paranasal CT imaging between 2020 and 2024 was analyzed. All diagnoses were confirmed histopathologically. Two representative CT slices per patient were selected at the patient level by an experienced radiologist blinded to final diagnosis. The dataset was split into 70…

By the Truvace desk1 min read1 sourcePeer-reviewedTRV-2026-1331· cite & verify
CT-based deep learning for identifying sinonasal inverted papilloma: a single-center retrospective study
PARANASAL SINUS CT · ANNOTATED VIDEO by Wilfredor. CC BY-SA 4.0 · https://creativecommons.org/licenses/by-sa/4.0

In brief

Researchers developed and tested CT-based deep learning models to distinguish sinonasal inverted papilloma from nasal polyps and normal anatomy using 871 retrospectively collected cases from 2020-2024. A ResNet50 convolutional network trained on two representative slices per patient achieved 98.47% test accuracy, outperforming a Vision Transformer model at 79.65%.

The result matters because inverted papilloma requires different management than polyps, and an imaging-based decision-support tool could improve triage before histopathology. Uncertainty remains about generalizability beyond this single center, performance with full volumetric CT and clinical variables, and prospective validation in routine workflows.

Main points

  1. Retrospective analysis of 871 patients who underwent paranasal CT imaging between 2020 and 2024 with histopathologically confirmed diagnoses.
  2. Two representative CT slices per patient selected at the patient level by an experienced radiologist blinded to final diagnosis.
  3. ResNet50 CNN achieved 98.47% test accuracy, outperforming Vision Transformer at 79.65%, with high class-wise performance for inverted papilloma detection.

The gain

A CT-based ResNet50 deep learning model achieved 98.47% test accuracy for classifying sinonasal inverted papilloma versus nasal polyps and normal anatomy, serving as a decision-support tool to assist clinical triage.

The rundown

The study used 871 patients from 2020-2024, split 70% training and 30% testing with 20% of training for validation, and compared ResNet50 against Vision Transformer. Selection was done at the patient level by a radiologist blinded to final diagnosis, and all diagnoses were histopathologically confirmed.

Authors position the model as complementing rather than replacing histopathology, proposing a hybrid diagnostic strategy. Performance was reported as high class-wise for inverted papilloma detection, with ResNet50 substantially outperforming the transformer model.

Sources

  1. Peer-reviewedEuropean Archives of Oto-Rhino-Laryngology2026-10-04

The debate