364P - Artificial intelligence provides more accurately neck lymph nodes auto-segmentation in radiotherapy

Background

To develop artificial intelligence auto-segmentation model that generates consistent, high-quality lymph nodes contouring in head and neck cancer patients who received radiotherapy.

Methods

There were 60 computed tomography (CT) scans were retrospectively selected into training and another 60 CT scans were collected into cross-validation. All target delineations covered head and neck lymph node level I through V and based on the Radiation Therapy Oncology Group (RTOG) guideline. All targets were approved by radiation oncologists specializing in head and neck cancer. The volume of interest and all approved contours were used to train a 3D U-Net model. Different lymph node levels were trained independently. The trained model was used on cross-validation group. Auto-segmentations were revised by 2 radiation oncologists.

Results

The Dice Similarity Coefficients were 0.79 and 0.88 in trained group and cross-validation group. The volume changes ranged from -22.2 to 89.0 cm³. The center shift for x-direction, y-direction, and z-direction were -0.57 to 0.16 cm, -0.14 to 0.88 cm, and -0.19 to 0.38 cm, respectively.

Conclusions

We developed an artificial intelligence auto-segmentation model to autodelineate head and neck lymph nodes. Most results of auto-segmentations were acceptable after radiation oncologist review. This enables more efficient and consistent targeting of neck lymph nodes in radiation treatment planning.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The author.

Funding

Has not received any funding.

Disclosure

The author has declared no conflicts of interest.