Photo Credit: freepik

The following is a summary of “Deep Learning-Assisted Interactive Contouring of Lung Cancer: Impact on Contouring Time and Consistency,” published in the September 2024 issue of Oncology by Trimpl et al.

This study investigates the effects of a deep learning (DL)-assisted interactive contouring tool on inter-observer variability and the time required for gross tumor volume (GTV) delineation in patients with non-small cell lung cancer (NSCLC). 

To evaluate the tool’s efficacy, nine clinicians contoured the GTVs using PET-CT scans of 10 patients with NSCLC, alternating between a DL-assisted tool and standard manual contouring. Each clinician contoured a case using one method and repeated the task with the alternate method one week later, allowing for a direct comparison of contours and time metrics between the two approaches. The results demonstrated that the DL-assisted tool significantly reduced the active contouring time by 23% compared to manual contouring (p < 0.01), cutting the average time per case from 22 minutes to 19 minutes. Additionally, while the average observation time remained consistent across both methods, accounting for approximately 60% of the interaction time, the DL-assisted tool notably decreased contour variability in tumor regions where clinicians typically showed the most disagreement. 

However, the overall consensus contours were consistent across both methods. These findings suggest that integrating a DL-assisted contouring tool into clinical practice could enhance the efficiency of tumor delineation and reduce inter-observer variability, particularly in challenging regions. This improvement in contouring efficiency and consistency could lead to more streamlined workflows and potentially better patient outcomes in radiation oncology.

Source: sciencedirect.com/science/article/pii/S0167814024007709