Comparison of Different Fusion Radiomics for Predicting Benign and Malignant Sacral Tumors: A Pilot Study

Fei Zheng ¹, Ping Yin ¹, Kewei Liang ²

⁰Department of Radiology, Peking University People's Hospital, No. 11 Xizhimen South Street, Xicheng District, Beijing, 100044, People's Republic of China.

Journal of Imaging Informatics in Medicine +

|

May 8, 2024

View abstract on PubMed

Summary

This study developed a deep learning radiomic nomogram (DLRN) to differentiate benign from malignant sacral tumors. The DLRN achieved high accuracy and AUC, offering a valuable tool for clinical decisions.

Area Of Science

Radiology
Oncology
Medical Imaging

Background

Accurate differentiation between benign and malignant sacral tumors is critical for treatment planning.
Sacral tumors pose diagnostic challenges due to their location and varied presentations.

Purpose Of The Study

To develop and evaluate a deep learning radiomic nomogram (DLRN) for distinguishing benign from malignant sacral tumors.
To compare the performance of deep learning (DL) and classical machine learning (CML) fusion models.

Main Methods

Retrospective review of axial T2-weighted imaging (T2WI) and non-contrast computed tomography (NCCT) from 134 patients with pathologically confirmed sacral tumors.
Development of two benchmark fusion models (DL and CML) using multi-modal imaging features.
Formulation of the DLRN by integrating the best-performing benchmark model with clinical data.

Main Results

The DL benchmark fusion model outperformed the CML fusion model.
The DLRN demonstrated superior predictive performance with an accuracy of 0.889 and an area under the receiver operating characteristic curve (AUC) of 0.961 in test sets.
Calibration curves and decision curve analysis (DCA) confirmed the DLRN's predictive capability and clinical utility.

Conclusions

The DLRN is a robust predictive tool for differentiating benign and malignant sacral tumors.
This model can aid in risk stratification and inform clinical treatment decisions, improving patient management.

Keywords:

Deep learning Multimodality imaging Nomogram Radiomics Sacral tumors