| Autor: |
Rahmad, Fakhrul Razan, Wan Zakaria, Wan Nurshazwani, Nazari, Ain, Tomari, Mohd Razali Md, Suberi, Anis Azwani, Nik Fuad, Nik Farhan |
| Předmět: |
|
| Zdroj: |
AIP Conference Proceedings; 2023, Vol. 2564 Issue 1, p1-8, 8p |
| Abstrakt: |
Ischemic stroke can happen from insufficiency of blood flow to the brain tissue due to the presence of thrombus in the blood vessel which obstruct blood flow to the brain. If left untreated, this condition can lead to necrosis of the brain tissue and stroke. Medical practitioners have been using non-contrast computed tomography (NECT) as an early diagnostic tool in visualizing patients' brain to identify ischemia. However, ischemic stroke from NECT sometimes is only slightly visible and can be very hard to perceive. Not only that, NECT may produce artifacts from the beam radiation and can affect the radiologists' perception when looking at the images. Although Magnetic Resonance Imaging (MRI) is a much more preferable tool when detecting ischemic stroke, it can be hard to come by at necessary time in addition to being a complicated process. Recent studies have shown multiple technique that can be utilized to help medical practitioners diagnose ischemic stroke from NECT images with the aid of medical imaging and deep learning (DL) to overcome the limitations of NECT. One of the most common way DL can be used to assist in medical field is telling apart ischemic from non-ischemic by mean of classification. This paper compares three networks; DarkNet-19, DarkNet-53 and ResNet-50 and assess the performance and processing time of each network. ResNet-50 is empirically proven to be the best with 100% validation frequency and short training period with 1.7286 second per image apart from yielding accuracy, precision and sensitivity of 92.5%, 95% and 0.9048 respectively. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
