Automatic Renal Cortical Defect Detection in Tc-99 DMSA Scintigraphy using Multi-view Deep Learning

• Main Authors: Hajianfar, G., Bagheri, S., Askari, E., Sabouri, M., Aghaei, A., Eyvazkhani, T., Rasouli, A., Shiri, I., Zaidi, H.
• Format: Conference Proceeding
• Language: English
• Subjects: Analytical models; Deep learning; Medical services; Microwave integrated circuits; Reliability; Semiconductor detectors; Semiconductor device modeling
• ISSN: 2577-0829
• DOI: 10.1109/NSSMICRTSD49126.2023.10337859

Renal dysfunction can arise from various factors, and detecting renal scarring (RS) is essential to prevent long-term consequences. DMSA scintigraphy is useful in detecting cortical anomalies and is considered the gold standard for diagnosing RS. Deep Learning (DL) has shown promising results in medical imaging, especially with multi-view input DL. This work aimed to develop a deep learning model that can automatically detect cortical defects in the kidney by analyzing multiple views of DMSA images. This study included 747 patients with DMSA images and eight different views to detect kidney cortical defects using deep learning. The ResNet152V2 model was applied to each view separately, followed by the Attention Augmented (AA) method. The model developed in this study showed promising results in detecting a cortical defect in DMSA scans with AUC=0.76, Sensitivity=0.84, and Specificity=0.63, which can be challenging for physicians. Furthermore, our model has shown promising results in automatically detecting cortical defects in the kidneys using DMSA images with multi-view input. This suggests that developing similar models could save time to physicians and be used as reliable assistants in diagnosing renal dysfunction.