An Inception based Urothelial Cell Classification Network for the detection of Bladder Carcinoma from Urine Cytology Microscopic Images

Medical image diagnostics now benefit greatly from the use of deep convolutional neural networks. The CNN-based deep neural networks are extensively used in the medical classification tasks. Although deep learning algorithms have gained performance comparable to pathologists in interpreting whole sl...

Full description

Saved in:
Bibliographic Details
Published in2023 International Conference on Control, Communication and Computing (ICCC) pp. 1 - 5
Main Authors NP, Arathy Menon, P.N., Pournami, P B, Jayaraj
Format Conference Proceeding
LanguageEnglish
Published IEEE 19.05.2023
Subjects
Bladder
Bladder cancer
Computer architecture
Deep learning
Microprocessors
Microscopy
neural network
Reliability
Task analysis
urine cytology
urothelial carcinoma
Online AccessGet full text

Cover

More Information
Summary:Medical image diagnostics now benefit greatly from the use of deep convolutional neural networks. The CNN-based deep neural networks are extensively used in the medical classification tasks. Although deep learning algorithms have gained performance comparable to pathologists in interpreting whole slide images for the detection of tumours, very few researchers have explored the possibility of detecting urothelial carcinoma from microscopic images. In this study, we propose a novel deep learning model for urine cytology smear-based detection of urothelial cell cancer (UCC). The network is based on Inception architecture that can efficiently learn the features of varied size cells in the image and produced relatively high accuracy when compared to state-of-the-art techniques. The proposed technique is evaluated using a dataset that includes the cytology samples of 115 individuals, 59 of whom had UCC instances that were histologically confirmed and the remaining 56 benign cases were identified through routine cytology samples. The suggested approach offers 98.63% accuracy with fewer parameters. The method's performance in terms of accuracy and parameter count is highly encouraging.
DOI:10.1109/ICCC57789.2023.10165205