Gaze-SPV: An End-to-End Approach for Enhancing Simulated Prosthetic Vision using Gaze for Object Recognition

Ashkan Nejad, Burcu Kucukoglu, Jaap de Ruyter van Steveninck, Marcel van Gerven

Neuroprosthetic implants using electrical neurostimulation of the visual pathway are promising technology for partially recovering vision in people with visual impairments. Recent methods were introduced for representation of the environment in prsthetic vision do not incorporate the gaze locations. In this project we aim to design an end-to-end deep learning method for the simulated vision representation using gaze locations.

ACE-DNV: Automatic Classification of Gaze Events in Dynamic Natural Viewing

Ashkan Nejad, Gera de Haan, Joost Heutink, Frans Cornelissen

It is challenging for people with visual field defects to perform daily tasks that rely on having a good visual overview. For helping people with such a condition, an essential step is to quantify their scanning behavior. However, there are no accurate gaze event detectors that are suitable for use in dynamic natural conditions, limiting research in such settings. ACE-DNV is one of the first eye-movement event classification methods in dynamic and natural viewing. Our method only uses recordings of commercial eye-trackers We aim to design a gaze-event detector for conditions that allow free head- and body- movements conditions.
Published

A Fast and Memory-efficient Brain MRI Segmentatin Framework for Clinical Applications

Ashkan Nejad, Saeed Masoudnia, Mohammad-Reza Nazem-Zadeh

Current structural brain MRI segmentation methods have limited use in the clinics due to their time and memory consumption. To address this issue, we customize a memory-efficient (GPU) brain structure segmentation framework based on DNN. The code is publicly available.
Published in 2022 44th IEEE EMBC

A Memory-efficient Deep Framework for Multi-Modal MRI-based Brain Tumor Segmentation

Nima Hashemi, Saeed Masoudnia, Ashkan Nejad, Mohammad-Reza Nazem-Zadeh

To address the memory limitations in automatic brain tumor segmentation, we utilize several techniques for customizing a memory-efficient yet accurate deep framework based on 2D U-nets. In this framework, the simultaneous multi-label tumor segmentation is decomposed into fusion of sequential binary segmentation tasks. Experiments on BraTS 2020 showed that our framework almost achieves state-of-the-art results. Dice scores of 0.905, 0.903, and 0.822 for whole tumor, tumor core, and enhancing tumor are accomplished on the testing set.
Published in 2022 44th IEEE EMBC

Saliency Map-based Image Retrieval using Krawtchouk Moments

Ashkan Nejad, Mohammad Reza Faraji, Xiaojun Qi

CBIR has been widely used to find the similar images based on the semantic meaning of the content. In this project at IASBS, we have focused on proposing a new method for Image Retrieval by using Saliency maps and Krawtchouk Moments. The supervisor of this project is Dr. Mohammadreza Faraji.
Under Review

Helia

Helia is a AI-based Electronic Health Recording System which assists physicians in the process of diagnosis and prescribing to prevent medical error. Helia uses a powerful AI engine with huge amount of data extracted from Medical Hand-books and experiences in other cases. IK Hospital, one of the biggest in Iran, has been the first to use this project. For more information please visit: helia-care.com