Oropharyngeal primary tumor segmentation for radiotherapy planning on magnetic resonance imaging using deep learning

Abstract

Background and purpose: Segmentation of oropharyngeal squamous cell carcinoma (OPSCC) is needed for radiotherapy planning. We aimed to segment the primary tumor for OPSCC on MRI using convolutional neural networks (CNNs). We investigated the effect of multiple MRI sequences as input and we proposed a semi-automatic approach for tumor segmentation that is expected to save time in the clinic.

Materials and methods: We included 171 OPSCC patients retrospectively from 2010 until 2015. For all patients the following MRI sequences were available: T1-weighted, T2-weighted and 3D T1-weighted after gadolinium injection. We trained a 3D UNet using the entire images and images with reduced context, considering only information within clipboxes around the tumor. We compared the performance using different combinations of MRI sequences as input. Finally, a semi-automatic approach by two human observers defining clipboxes around the tumor was tested. Segmentation performance was measured with Sørensen-Dice coefficient (Dice), 95th Hausdorff distance (HD) and Mean Surface Distance (MSD).

Results: The 3D UNet trained with full context and all sequences as input yielded a median Dice of 0.55, HD of 8.7 mm and MSD of 2.7 mm. Combining all MRI sequences was better than using single sequences. The semi-automatic approach with all sequences as input yielded significantly better performance (p < 0.001): a median Dice of 0.74, HD of 4.6 mm and MSD of 1.2 mm.

Conclusion: Reducing the amount of context around the tumor and combining multiple MRI sequences improved the segmentation performance. A semi-automatic approach was accurate and clinically feasible.

Keywords: Convolutional neural network; MRI; Oropharyngeal cancer; Radiotherapy; Segmentation; Semi-automatic approach.

Fig. 1

Original MRI image with the manual segmentation (green) of the oropharyngeal tumor. The blue boxes are the bounding boxes of the tumor. The rest of the boxes are used as inputs to the network during training. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2

Segmentation performance in terms of Dice, 95th HD and MSD for the 3D. The different boxes show different MRI sequences as input: T1w (T1 weighted), T2w (T2 weighted), T1gd (T1 3D after gadolinium injection), T1gd and T2w (T1 3D after gadolinium injection and T2 weighted) and combining all sequences (All). The box includes points within the interquartile range (IQR) while the whiskers show points within 1.5 times the IQR.

Fig. 3

Segmentation performance of the semi-automatic approach with boxes drawn by two human observers. We compare the semi-automatic results (Ob 1 and Ob 2) to the fully automatic approach (Full). The box includes points within the interquartile range (IQR) while the whiskers show points within 1.5 times the IQR. Significance is represented as one star (*) for p < 0.01 and two stars (**) for p < 0.001.

Fig. 4

Robustness analysis. Segmentation performance in terms of median Dice, 95th HD and MSD for the semi-automatic approach as a function of the tumor centroid displacement and the clipbox diagonal length difference. The grey areas correspond to undetermined values due to the geometric constraints (i.e. no combination of shifts can achieve those values of centroid displacement and diagonal length difference).

Fig. 5

Comparison of the oropharyngeal segmentations in three different patients (a, b, c) trained with the fully automatic approach (red), with the semi-automatic approach (blue) and the manual delineation (green). The yellow boxes are the boxes drawn by the observer. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)