AUTHOR=Sethanan Kanchana , Pitakaso Rapeepan , Srichok Thanatkij , Khonjun Surajet , Weerayuth Nantawatana , Prasitpuriprecha Chutinun , Preeprem Thanawadee , Jantama Sirima Suvarnakuta , Gonwirat Sarayut , Enkvetchakul Prem , Kaewta Chutchai , Nanthasamroeng Natthapong 

TITLE=Computer-aided diagnosis using embedded ensemble deep learning for multiclass drug-resistant tuberculosis classification

JOURNAL=Frontiers in Medicine

VOLUME=Volume 10 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/medicine/articles/10.3389/fmed.2023.1122222

DOI=10.3389/fmed.2023.1122222

ISSN=2296-858X

ABSTRACT=In this study aims to develop the web application name Tuberculosis Drug Resistant Diagnosis System-CXR (TB-DRD-CXR), which is used to categorize TB patients into subgroups based on their level of drug resistance. The web application employs an ensemble deep learning model to classify tuberculosis strains into five subtypes: drug sensitive tuberculosis (DS-TB), drug resistant TB (DR-TB), multidrug-resistant TB (MDR-TB), pre-extensively drug-resistant TB (pre-XDR-TB), and extensively drug-resistant TB (XDR-TB). In addition to novel fusion techniques, the ensemble deep learning model proposed here includes image segmentation, data augmentation, and several learning rate strategies. The proposed model was compared to both state-of-the-art techniques and the standard homogeneous CNN architectures reported in the literature. The computational results indicate that the suggested method provides 4.0%-33.9% more accurate than the existing methods found in the literatures. The proposed model can also increase solution quality when compared to the standard CNN model such as DenseNet201, NASNetMobile, EfficientNetB7, EfficientNetV2B3, EfficientNetV2M, and ConvNeXtSmal by 28.8%, 93.4%, 2.99%, 48.0%, 4.4%, and 7.6%, respectively. The TB-DRD-CXR online application was developed and tested with 33 medical staff. Computational results revealed that it has a 96.7% accuracy rate found, a time-based efficiency (ET) of 4.16 goals/minutes, and an overall relative efficiency (ORE) of 100 percent. The system usability scale (SUS) of the proposed application is 96.7%. This means that the TB-DRD-CXR is the application that users will like using and will likely recommend to others (standard judgement from previous literature using SUS score).