A transformer-based multi-task deep learning model for simultaneous infiltrated brain area identification and...

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate identification of glioma-infiltrated brain areas and tumor boundaries is crucial for treatment planning.
• Manual delineation of gliomas is time-consuming and labor-intensive.
• Existing deep learning models primarily focus on tumor segmentation or feature prediction, with limited attention to infiltrated areas.

Purpose of the Study:

• To develop a novel deep learning model for simultaneous identification of infiltrated brain areas and glioma segmentation.
• To improve the efficiency and accuracy of neuro-oncological assessments.

Main Methods:

• A transformer-based multi-task deep learning model was developed.
• The model simultaneously performs infiltrated brain area identification and glioma segmentation.
• A retrospective study of 354 glioma patients (grades II-IV) was conducted, with data split into training, validation, and independent test sets.

Main Results:

• The multi-task model achieved an AUC of 94.95% on the independent test set.
• High sensitivity (87.67%), specificity (87.31%), and accuracy (87.41%) were reported for infiltrated area identification.
• A mean Dice score of 87.60% was achieved for whole tumor segmentation.

Conclusions:

• The developed multi-task model demonstrates superior performance compared to state-of-the-art methods.
• The model shows significant potential as an innovative solution for identifying tumor-infiltrated brain areas.
• This tool can practically support clinical decision-making in glioma management.