Computer-Aided Grading of Gliomas Combining Automatic Segmentation and Radiomics

Area of Science:

• Neuro-oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Gliomas are primary brain tumors requiring accurate grading for effective treatment.
• Objective grading of gliomas is crucial but can be challenging.

Purpose of the Study:

• To develop and evaluate an automatic computer-aided diagnosis (CAD) system for glioma grading.
• To combine automatic segmentation and radiomics for improved diagnostic accuracy.

Main Methods:

• A multiscale 3D convolutional neural network was used for automatic tumor segmentation from MRI data.
• Radiomic features (first-order, shape, texture) were extracted.
• Support vector machines and extreme gradient boosting with cross-validation were employed for feature selection and classification.

Main Results:

• The CAD system demonstrated high effectiveness in grading gliomas.
• Achieved an accuracy of 91.27%, weighted macroprecision of 91.27%, weighted macrorecall of 91.27%, and weighted macro-F1 score of 90.64% on a dataset of 274 gliomas.
• The system was evaluated on 220 high-grade and 54 low-grade gliomas.

Conclusions:

• The proposed CAD system shows significant potential for clinical application in assisting radiologists.
• The integration of automatic segmentation and radiomics enhances diagnostic capabilities for glioma grading.
• The system offers a highly accurate and automated approach to a critical aspect of neuro-oncology.