DeepCERES: A deep learning method for cerebellar lobule segmentation using ultra-high resolution multimodal MRI

Sergio Morell-Ortega, Marina Ruiz-Perez, Marien Gadea, Roberto Vivó Hernando, Gregorio Rubio, Fernando Aparici-Robles, María de la Iglesia-Vayá, Gwenaelle Catheline, Boris Mansencal, Pierrick Coupé, José V. Manjón

Abstract: Highlights•An ultra-high resolution multimodal cerebellar segmentation method.•A novel architecture named DPN that complements classic U-Net model.•A combination of multi-atlas techniques with Deep Learning to enhance method robustness.•Validation of cascade-based approach, utilizing ensemble techniques, to improve model performance.•A new free online pipeline to segment standard resolution T1 images is also proposed.