Publications

@conference{Souid20231136,

title = {Full Interpretability CBMIR to Help Minimize Radiologist Analysis Search Time},

author = {A. Souid and S. B. Othman and M. Hamroun and H. Sakli},

url = {https://www.scopus.com/inward/record.uri?eid=2-s2.0-85167737946\&doi=10.1109%2fIWCMC58020.2023.10183105\&partnerID=40\&md5=e4d8ee981345e84ba778805d96cfe8d9},

doi = {10.1109/IWCMC58020.2023.10183105},

isbn = {9798350333398},

year  = {2023},

date = {2023-01-01},

journal = {2023 International Wireless Communications and Mobile Computing, IWCMC 2023},

pages = {1136-1141},

publisher = {Institute of Electrical and Electronics Engineers Inc.},

abstract = {The area of medical imaging has experienced a remarkable expansion in recent times, with radiologists heavily depending on medical images for precise diagnoses. However, radiologists often encounter difficulties when presented with cases of uncertain diagnoses as they struggle to find comparable cases in both public and internal databases to assist them in their decision-making. This manual search process not only detracts from the ability to diagnose new cases but also places a considerable burden on radiologists' workflow. To tackle these challenges, this study suggests a computer-aided system that can automatically retrieve annotated medical images with similar content. This system incorporates a feature extractor based on deep learning, designed to simplify the process of identifying comparable images from an extensive chest radiograph database. The main focus of this study is on the feature extractor, which utilizes YOLOv5 and EfficientNet. YOLOv5 is known for its fast and efficient object detection framework, while the EfficientNet-B0 encoder block serves as an image retrieval and EfficientNet-B3 helps prevent noisy predictions. The study conducted experiments on a large dataset to evaluate the performance of the proposed solution. The results demonstrate that the model outperforms other classical approaches, with a mean average precision of 0.488 at a threshold of 0.9, which is higher than the performance of YOLOv5+ResNet, and EfficientDet of 0.234 and 0.257 respectively. Moreover, the model precision was significantly higher, with a precision of 0.864 for all pathologies, which was 0.352 higher than YOLOv5+ResNet and EfficientDet. Overall, the proposed computer-assisted system has the potential to enhance the efficiency of radiologists' workflow by offering a more effective and efficient tool for retrieving similar annotated medical images. This study is a significant advancement in the development of intelligent and efficient image retrieval systems in the medical imaging field. The proposed solution has the potential to significantly impact the diagnosis of complex medical cases, particularly in emergency departments. © 2023 IEEE.},

note = {cited By 1},

keywords = {CBMIR; Deep learning; Feature extractor; Interpretability; Lung pathology; Performance; Single stage; Single-stage detection; Work-flows; X-ray analyse, Computer aided analysis; Computer aided instruction; Decision making; Deep learning; Diagnosis; Energy dispersive X ray analysis; Image annotation; Image enhancement; Image retrieval; Large dataset; Object detection; Pathology; Search engines, Medical imaging},

pubstate = {published},

tppubtype = {conference}

}