MRI-based automatic segmentation of rectal cancer using 2D U-Net on two independent cohorts
Franziska Knuth, Ingvild Askim Adde, Bao Ngoc Huynh, Aurora Rosvoll Groendahl, René Mario Winter, Anne Negård, Stein Harald Holmedal, Sebastian Meltzer, Anne Hansen Ree, Kjersti Flatmark, Svein Dueland, Knut Håkon Hole, Therese Seierstad, Kathrine Røe Redalen, Cecilia Marie Futsaether
Abstract
BACKGROUND: Tumor delineation is time- and labor-intensive and prone to inter- and intraobserver variations. Magnetic resonance imaging (MRI) provides good soft tissue contrast, and functional MRI captures tissue properties that may be valuable for tumor delineation. We explored MRI-based automatic segmentation of rectal cancer using a deep learning (DL) approach. We first investigated potential improvements when including both anatomical T2-weighted (T2w) MRI and diffusion-weighted MR images (DWI). Secondly, we investigated generalizability by including a second, independent cohort. MATERIAL AND METHODS: ) as performance measure. The optimized models were evaluated on a C1 hold-out test set and the generalizability was investigated using C2. RESULTS: of 0.59. CONCLUSION: T2w MR-based DL models demonstrated high performance for automatic tumor segmentation, at the same level as published data on interobserver variation. DWI did not improve results further. Using DL models on unseen cohorts requires caution, and one cannot expect the same performance.