ESTRO 2024 - Abstract Book

S4441

Physics - Machine learning models and clinical applications

ESTRO 2024

Material/Methods:

The dataset comprised 117 patients treated with TMI/TMLI between 2011-2023 at IRCCS Humanitas Research Hospital. For each patient, the computed tomography (CT), RT structures, and RT plan were exported in DICOM format. The CNN input image consisted of three channels, obtained by projecting along the sagittal plane: (i) average CT pixel intensity within the PTV; (ii) PTV mask; (iii) brain, lungs, liver, bowel, and bladder masks. This “averaged” frontal view was built to aggregate the information analyzed by the MP when setting the field geometry in the TPS. Two CNNs were trained to predict the isocenters coordinates and jaws apertures for patients with (CNN-1) and without (CNN-2) isocenters on the arms. Given that slight changes in field geometry may still be acceptable, model evaluation was performed on a test set of 15 patients in two ways: (i) by computing the root mean squared error (RMSE) between the CNN output and ground truth; (ii) with a qualitative assessment of manual and generated field geometries - scale: 1 = not adequate, 4 = adequate - carried out in blind mode by three MPs with different expertise in TMI/TMLI. The Wilcoxon signed-rank test was used to evaluate for each MP the independence of the given scores between manual and automatic configurations, with significance level set at p<0.05.

Results:

The average and standard deviation values of RMSE for CNN-1 and CNN-2 were 13±3 mm and 18±4 mm, respectively. For the qualitative evaluation, the CNNs were integrated into a planning automation software for TMI/TMLI such that the MPs could analyze in detail the proposed field geometries directly in the TPS [2]. Figure 1 shows the CNN input image, and the manual and generated field geometries for two representative cases with and without isocenters on the arms. The selection of the CNN model to create the field geometry was based on the PTV width: in case the PTV was larger than 47.5 cm, then CNN-1 was used. This approach was employed to approximate the decision process of an experienced MP and provide a single option of field configuration.