Tinnitus classification based on resting-state functional connectivity using a convolutional neural network...

Area of Science:

• Neuroimaging
• Machine Learning
• Medical Diagnostics

Background:

• Aberrant functional connectivity (FC) in subjective tinnitus patients has been widely studied using resting-state functional MRI (rs-fMRI).
• The diagnostic efficacy of rs-fMRI FC as a biomarker for tinnitus remains unverified.
• Developing objective diagnostic tools for subjective tinnitus is crucial for clinical practice.

Purpose of the Study:

• To establish a convolutional neural network (CNN) model for distinguishing tinnitus patients from healthy controls based on rs-fMRI FC.
• To evaluate the diagnostic value of rs-fMRI FC as an imaging marker for subjective tinnitus.
• To provide guidance and fast diagnostic tools for the clinical diagnosis of subjective tinnitus.

Main Methods:

• A CNN model was trained on rs-fMRI data from 100 tinnitus patients and 100 healthy controls.
• An asymmetric convolutional layer was utilized within the CNN architecture.
• The CNN model was compared against traditional machine learning and transfer learning models, with all models tested on three different brain atlases.

Main Results:

• The CNN model demonstrated superior performance compared to traditional and transfer learning models, achieving the highest area under the curve (AUC = 0.944) on the Dos_160 atlas.
• The best-performing model identified the default mode network, salience network, and sensorimotor network as critical in differentiating tinnitus patients from controls.
• The results highlight the potential of specific brain networks in the objective diagnosis of subjective tinnitus.

Conclusions:

• The developed CNN model accurately diagnoses tinnitus patients using rs-fMRI data.
• This study confirms the diagnostic value of functional connectivity measured by rs-fMRI for subjective tinnitus.
• The findings support the use of rs-fMRI FC and CNNs as a potential clinical diagnostic tool.