Classification and Visualization of Chemotherapy-Induced Cognitive Impairment in Volumetric Convolutional Neural...

Area of Science:

• Neuroscience
• Oncology
• Artificial Intelligence

Background:

• Breast cancer is the most common female cancer globally, with chemotherapy survivors often experiencing cognitive impairment, termed "chemo-brain."
• Chemotherapy-induced cognitive impairment (CICI) significantly impacts survivors' quality of life, yet objective detection methods are limited.
• Functional MRI (fMRI) combined with deep learning offers a potential avenue for identifying CICI.

Purpose of the Study:

• To develop and validate deep learning models for detecting cerebral alterations indicative of chemo-brain in breast cancer survivors.
• To identify specific brain regions and functional patterns associated with chemotherapy-induced cognitive impairment using fMRI data.
• To establish a foundation for future clinical tools to track chemo-brain.

Main Methods:

• Recruited 55 post-chemotherapy breast cancer survivors and 65 healthy controls.
• Extracted mean fractional amplitudes of low-frequency fluctuations (mfALFF) from resting-state fMRI as input features.
• Utilized 3D-transformed ResNet-50 and DenseNet-121 deep learning models, enhanced with squeeze and excitation (SE) blocks, to differentiate between groups. Integrated gradients were used for visualization.

Main Results:

• Both SE-ResNet-50 and SE-DenseNet-121 models achieved an average accuracy of 80% in differentiating chemo-brain from healthy controls.
• SE-DenseNet-121 demonstrated higher precision (86%) and recall (80%) compared to SE-ResNet-50 (78% precision, 70% recall).
• Integrated gradients highlighted the frontal, temporal, parietal, and occipital lobes as critical regions for identifying chemo-brain, correlating with default mode and dorsal attention networks.

Conclusions:

• Deep learning models effectively identify chemo-brain using fMRI-derived features, offering a promising approach for objective assessment.
• The identified brain regions and networks provide insights into the neural underpinnings of chemotherapy-induced cognitive impairment.
• These findings pave the way for developing advanced clinical tools for monitoring chemo-brain in breast cancer survivors.