Hippocampus segmentation on epilepsy and Alzheimer's disease studies with multiple convolutional neural networks

• DOI: 10.1016/j.heliyon.2021.e06226
• Publication Date: 2021-02-10T14:42:56Z
• Authors: Diedre Carmo; Bruna Silva; Clarissa Yasuda; Letícia Rittner; Roberto Lotufo
• Keywords: Dice

Background: Hippocampus segmentation on magnetic resonance imaging is of key importance for the diagnosis, treatment decision and investigation neuropsychiatric disorders. Automatic an active research field, with many recent models using deep learning. Most current state-of-the art hippocampus methods train their healthy or Alzheimer's disease patients from public datasets. This raises question whether these are capable recognizing a different domain, that epilepsy resection.New Method: In this paper we present state-of-the-art, open source, ready-to-use, learning based method. It uses extended 2D multi-orientation approach, automatic pre-processing orientation alignment. The methodology was developed validated HarP, dataset.Results Comparisons: We test alongside other methods, in two domains: HarP set in-house dataset, containing resections, named HCUnicamp. show our method, while trained only surpasses others literature both HCUnicamp Dice. Additionally, Results training testing volumes also reported separately, comparisons between data vice versa.Conclusion: Although state-of-the-art including own, achieve upwards 0.9 Dice all tested produced false positives resection regions, showing there still room improvement when involved.