Building a GUI Tool for Automated Aortic Segmentation in Low-Dose Chest CT Images with PET-Based Standard Uptake...

The use of PET-CT combined with Fluorodeoxyglucose (FDG) has enhanced the detection and management of conditions like vasculitis and lymphoma. However, the current approach to measuring standardized uptake value (SUV) requires manual selection of Regions of Interest (ROI), which is time-intensive. This study introduces an automated method to measure tracer uptake in the aorta, aiming to improve efficiency and accuracy.

We conducted a proof-of-concept study on PET-CT scans, including those with vasculitis, lymphoma, and healthy controls. A UNET model with a ResNet-18 backbone was used for automated aorta segmentation and SUVmax calculation. We assessed the model's performance using the Intersection over Union (IoU) and compared SUVmax results from automated and manual methods.

The model achieved an IoU score of 0.905 in the validation slides set, reflecting strong segmentation performance. The mean SUVmax between automated and manual methods showed no significant difference (2.75 vs 2.28, p = 0.38). The automated process significantly reduced segmentation time from 36.3 minutes to 4.8 minutes.

Automated aortic segmentation and SUV calculation using deep learning techniques can enhance PET-CT diagnostic workflows by matching the accuracy of manual methods while greatly reducing processing time.