CervixMed: Detecting cervical cancer based on combinational data using hybrid architecture

Debashis Gupta, Aditi Golder, Md Mahfuzul Haque, Mohammad Ali Moni

Research output: Book chapter/Published conference paperConference paperpeer-review

Abstract

Cervical cancer is women's second most frequent malignancy globally, with a 60% mortality rate. Early identification through routine checkups is crucial because cervical cancer has a long latent period and begins with no overt signs. One of the most evident barriers to not having an early detection is the time required for the diagnostic process, which also accounts for indifference regarding its automation. These uncertainties occur for several reasons, while one of the most prominent causes is providing the report by the computer-aided diagnostic process only based on the patient's information in tabular form or only predicting the patient's cervix image. In this study, we proposed three concepts 1. Hybrid Architecture, 2. CervixMed, a novel CNN-based architecture, and 3. GUI (Graphical User Interface). In Hybrid Architecture, we propose a methodology that takes both types of data, i.e., tabular and image, and then merges the prediction provided by the models and shows the final output. For tabular data classification, we use traditional-machine learning models like KNN, SVM, LR, and RF, where LR and SVM provided the best accuracy of 99%. On the other hand, the image data is classified into one of the five categories using our novel lightweight CNN-based architecture, CervixMed, that outperforms other pre-trained models like DenseNet121, DenseNet169, DenseNet201, ResNetl52v2, and VGG19 by 94% in accuracy and 98% in sensitivity. The hybrid architecture combines the abovementioned models and chooses the best pair regarding accuracy. In this study, we demonstrated that combining LR and SVM with CervixMed outperforms the other combinations by 96.5% accuracy. Lastly, we demonstrated the methodology of the proposed GUI for interacting with the end-users with this hybrid model.

Original languageEnglish
Title of host publication2023 International Conference on Digital Image Computing: Techniques and Applications, DICTA 2023
PublisherIEEE, Institute of Electrical and Electronics Engineers
Pages570-577
Number of pages8
ISBN (Electronic)9798350382204
ISBN (Print)9798350382211
DOIs
Publication statusPublished - 2023
EventThe International Conference on Digital Image Computing: Techniques and Applications: DICTA 2023 - Sails Port Macquarie, Port Macquarie, Australia
Duration: 28 Nov 202301 Dec 2023
https://www.dictaconference.org/
https://www.dictaconference.org/?page_id=2623 (Conference program)

Conference

ConferenceThe International Conference on Digital Image Computing: Techniques and Applications
Country/TerritoryAustralia
CityPort Macquarie
Period28/11/2301/12/23
OtherDigital Image Computing: Techniques and Applications (DICTA) is the main Australian Conference on computer vision, image processing, pattern recognition, and related areas. DICTA was established in 1991 as the premier conference of the Australian Pattern Recognition Society (APRS).
DICTA provides a forum for researchers, engineers, and practitioners to present their latest findings and innovations in these areas, as well as to exchange ideas and discuss emerging trends and challenges in the field. The conference covers a wide range of topics, including image and video processing, machine learning, pattern recognition, and computer graphics, among others.​
Internet address

Fingerprint

Dive into the research topics of 'CervixMed: Detecting cervical cancer based on combinational data using hybrid architecture'. Together they form a unique fingerprint.

Cite this