Explainable CBMIR features extractor module to ease Thoracic aortic aneurysm retrieving preventing potential Acute Aortic Dissection | IEEE Conference Publication | IEEE Xplore

Explainable CBMIR features extractor module to ease Thoracic aortic aneurysm retrieving preventing potential Acute Aortic Dissection


Abstract:

Medical imaging plays an increasingly crucial role in radiology, facilitating accurate and efficient diagnosis of complex medical cases. One such case that requires atten...Show More

Abstract:

Medical imaging plays an increasingly crucial role in radiology, facilitating accurate and efficient diagnosis of complex medical cases. One such case that requires attention is aortic aneurysm, which often lacks noticeable symptoms and can result in sudden death if left undetected. In order to address this issue, we have developed a feature extractor block utilizing YOLOv5 and EfficientNet. This block effectively extracts and localizes thoracic aortic enlargement in chest radiography, enabling precise identification of the condition. Our research employs a two-stage training strategy, yielding commendable results for the abnormality and detection modules. The sensitivity and specificity achieved were 0.842 and 0.977, respectively. Additionally, the mean average precision at a threshold of 0.5 reached an average value of 0.884, while TAD (thoracic aortic enlargement) detection achieved a score of 0.994. Other metrics demonstrate favorable outcomes when compared to relevant literature. Consequently, this study represents a significant advancement towards the development of intelligent and efficient image retrieval systems, which have the potential to greatly impact the diagnosis of complex medical cases in the emergency department, specifically those related to TAD.
Date of Conference: 20-23 September 2023
Date Added to IEEE Xplore: 22 November 2023
ISBN Information:

ISSN Information:

Conference Location: Hammamet, Tunisia

Contact IEEE to Subscribe

References

References is not available for this document.