This repository contains data related to "Deep learning-based detection of optic-chiasmal malformation" publication, specifically the 5 pickle files containing the parameteres for the trained Convolutional Neural Networks (CNNs) based on the the 3D U-Net architecture.
### CNN weights
Each file containts a complete set of parameters that can be loaded into previously defined CNN architecture (see https://github.com/rjpuzniak/Use-of-deep-learning-based-optic-chiasm-segmentation-for-investigating-visual-system-pathophysiology) in order to obtain a CNN trained to segment the optic chiasm from anatomical (T1-weighted) MRI images of control participants.
The provided sets of parameters were generated during several CNN training sessions, differing in number of total epochs `ep` and set learning rate `lr`, where from each session a single set, achieving the best performance on the validation dataset was chosen. Each set is named accordingly in a format `{}ep_{}lr`, where `ep` corresponds to total number of epochs and `lr` the learning rate.
#### Usage
The files are needed to be downloaded and saved to the `1_Data/0_CNN_weights` folder. They will be later required by the scripts provided in the `2_Code/4_Deploy_CNN`, which will allow for loading chosen parameters into CNN and deployment of the CNN on the T1w images.