Naslov
Diagnostic of Bladder Cancer Using Hybrid Neural Networks Based on Edge Detectors

Autori
Ivan Lorencin, Josip Španjol, Antun Gršković, Zlatan Car

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
My First Conference Book of Abstracts / - Rijeka, 2020, 22-23

Skup
4th edition of annual conference for doctoral students of engineering and technology „MY FIRST CONFERENCE“

Mjesto i datum
Rijeka, Hrvatska, 24.09.2020

Vrsta sudjelovanja
Predavanje

Vrsta recenzije
Recenziran

Ključne riječi
Artificial Neural Network, AUC, Bladder Cancer Diagnosis, Edge detector, Hybrid model

Sažetak
Bladder cancer is one of the most common malignant diseases of the urinary tract and is the fourth most common malignant disease in men in Croatia [1]. The diagnostic procedure for bladder cancer usually consists of a biopsy and pathohistological findings. Such an approach can often be invasive and time consuming [2]. For this reason, an endomicroscopic method based on confocal laser endomicroscopy (CLE) supported by artificial intelligence algorithms is being introduced into clinical practice. The application of artificial intelligence in problems of medical image recognition is most often based on the application of artificial neural networks (ANN), most often convolutional neural networks (CNN). The selection of CNN models may require considerable computing resources, which are often unavailable in clinical practice. For this reason, edge detectorsbased neural network hybrid models are being introduced. Such approaches offer a stabile classification performance with much simpler neural network architectures [3]. In this paper, a multi-class classification approach that is based on four classes (high- grade carcinoma, low-grade carcinoma, carcinoma in situ and healthy mucosa) is presented. From obtained results it can be noticed that such an approach offers stabile classification performances for multi-class classification as well, achieving macro AUC of 0.98 and micro AUC of 0.97

Izvorni jezik
Hrvatski

Znanstvena područja
Elektrotehnika, Računarstvo, Temeljne medicinske znanosti

POVEZANOST RADA