Self-Supervised Anomaly Detection in Brain MRI Using Convolutional Autoencoders and Masked Autoencoders
Author(s) : Niamatellah Lahkim
I clarified how many slices were used for training and testing, and I explained why only T1 images were selected. I added more details about preprocessing, the choice of models, and how the evaluation metrics were computed. I also improved the comparison between the CNN autoencoder and the MAE by explaining which metrics were used to judge “better localisation.” Two example figures were added: one showing the MRI slice and another showing the error inside the tumour. Finally, I rewrote some sentences to remove repetition and make the text clearer and more consistent.