Automatic segmentation of kidneys in computed tomography images using U-Net

Area of Science:

• Medical Imaging
• Radiation Oncology
• Artificial Intelligence

Background:

• Accurate segmentation of organs at risk in CT images is crucial for radiation therapy planning.
• Manual segmentation is time-consuming, prone to inter-observer variability, and dependent on clinician experience.
• Deep learning (DL) offers a potential solution for automated segmentation.

Purpose of the Study:

• To utilize a DL-based method for segmenting kidneys in CT images.
• To facilitate radiotherapy treatment planning through automated kidney segmentation.
• To evaluate the efficacy of DL in improving segmentation accuracy and speed.

Main Methods:

• CT scans from 20 patients were analyzed.
• The U-Net model was employed for kidney segmentation.
• Quantitative evaluation used Dice Similarity Coefficient (DSC), Matthews Correlation Coefficient (MCC), Hausdorff Distance (HD), sensitivity, and specificity.

Main Results:

• The U-Net model demonstrated good accuracy in segmenting kidneys.
• Detailed performance metrics for kidney segmentation were obtained and presented.
• Results were benchmarked against recent findings from other studies.

Conclusions:

• Fully automated DL-based segmentation of CT images can significantly enhance radiotherapy organ contouring.
• This approach has the potential to improve both the speed and accuracy of the segmentation process.
• DL methods represent a promising advancement for radiotherapy treatment planning.