Prediction of lower-grade glioma molecular subtypes using deep learning

Area of Science:

• Neuro-oncology
• Medical imaging analysis
• Artificial intelligence in medicine

Background:

• Accurate molecular subtyping of lower-grade gliomas (LGG) is crucial for effective treatment planning.
• Preoperative diagnosis of LGG molecular subtypes remains a clinical challenge.
• Current diagnostic methods can be invasive and time-consuming.

Purpose of the Study:

• To develop and evaluate a deep learning model for preoperative, direct 3-group molecular subtyping of LGG.
• To utilize multimodal imaging data (MRI, PET, CT) for improved diagnostic accuracy.
• To establish a non-invasive method for predicting LGG molecular subtypes.

Main Methods:

• A deep learning model was designed to predict the 3-group molecular subtype of LGG.
• The model integrated multimodal data: Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET), and Computed Tomography (CT).
• Performance was assessed using leave-one-out cross-validation on a dataset of 217 LGG patients.

Main Results:

• The model achieved the highest accuracy (68.7%) when using combined MRI, PET, and CT data.
• The deep learning model's accuracy (68.7%) surpassed the conventional sequential method (65.9%) for predicting IDH and 1p/19q status.
• Individual imaging modalities yielded lower test accuracies: MRI (58.5%), MRI+PET (60.4%), and MRI+CT (59.4%).

Conclusions:

• A novel deep learning model enables direct, preoperative prediction of the 3-group LGG molecular subtype using multimodal imaging.
• The model demonstrated a high test accuracy of 68.7%, significantly improving upon conventional methods.
• This represents a breakthrough, doubling the expected accuracy for this classification task and offering a non-invasive diagnostic tool.