AI holds potential to differentiate between normal and abnormal kidneys in children on SPECT imaging

In a test, the model developed by C. Lin, Chang Gung Memorial Hospital, Taoyuan, Taiwan, and colleagues, achieved an almost perfect agreement with reads by two experienced experts.

Experimental results indicate that the AI (artificial intelligence) model based on SPECT images has the potential to help nuclear medicine physicians identify kidney defects in children.

"Our findings suggest that deep learning may be used as a computer-aided diagnosis system that assists less-experienced nuclear medicine physicians in the diagnosis of renal cortical defects," the researchers wrote. "Thus, they suggested an AI model that can differentiate between normal and abnormal kidneys "may be of clinical interest" in these patients.

The team investigated the feasibility of using deep learning to differentiate normal from abnormal (or scarred) kidneys using technetium-99m dimercaptosuccinic acid (99mTc-DMSA) single-photon-emission computed tomography (SPECT) in paediatric patients.

For this purpose, the researchers retrospectively reviewed three hundred and one 99mTc-DMSA renal SPECT examinations. 301 patients were randomly split into 261, 20, and 20 for training, validation, and testing data, respectively.

Training of the DL model was done using 3D SPECT images, 2D maximum intensity projections (MIPs), and 2.5-dimensional (2.5D) MIPs (i.e., transverse, sagittal, and coronal views). Each DL model was trained to determine SPECT images as either abnormal or normal. The reference standard was the consensus reading results by two nuclear medicine physicians.

Key findings include:

· The DL model trained by 2.5D MIPs outperformed that trained by either 3D SPECT images or 2D MIPs.

· The accuracy, sensitivity, and specificity of the 2.5D model for the differentiation between normal and abnormal kidneys were 92.5%, 90% and 95%, respectively.

· Further analysis revealed a kappa coefficient of 0.85 obtained from the 2.5D model, indicating that it provided an almost perfect agreement with the visual reading results of the two experienced physicians.

The findings were promising, despite this the study was limited by a small number of patient images. They suggested that such DL models ultimately should also apply to adults, including renal transplant donor candidates. They wrote that additional research is warranted.

"The study is the first using DL with Tc-99m DMSA SPECT images to evaluate differentiating normal from abnormal kidneys," the researchers noted.

Reference:

Lin, C., Chang, Y., Chiu, H., Cheng, C., & Huang, H. (2023). Differentiation between normal and abnormal kidneys using 99mTc-DMSA SPECT with deep learning in paediatric patients. Clinical Radiology, 78(8), 584-589. https://doi.org/10.1016/j.crad.2023.04.015