Automated Classification and Localization of Inner Ear Structures in MRI Using Deep Learning

Automated Classification and Localization of Inner Ear Structures in MRI Using Deep Learning

🎓 Program: Master in Machine Learning for Health
🏫 Institution: Universidad Carlos III de Madrid
🤝 In collaboration with: Clínica Universidad de Navarra
👩‍🎓 Student: Claudia Castrillón Álvarez
📆 Academic Year: 2024/2025
📝 Grade: 9.5 / 10
🔗 Related thesis: Laura Rodrigo Muñoz – Segmentation of Inner Ear Structures

This Master Thesis represents the first stage in a comprehensive pipeline for the automatic diagnosis of Ménière’s Disease from MRI. The focus was on the automated classification and localization of inner ear structures using two high-resolution imaging modalities: 3D-SPACE-MRC and 3D-REAL-IR MRI.

The system included:

• A custom Convolutional Neural Network to classify relevant MRI slices containing ear structures
• A YOLO-based object detection model to localize cochlear and vestibular regions

🧪 Highlights

• Classification model achieved 97.74% accuracy with low validation loss
• Object detector delivered strong performance in mAP@0.5 and mAP@[0.5:0.95]
• Dataset: MRI volumes from 90 patients, labeled in collaboration with radiologists
• The pipeline is designed for clinical integration into Siemens MRI systems

This work laid the foundation for the segmentation and volumetric analysis carried out in the paired thesis project.