Evaluation of a Deep Learning-Based Auto-Segmentation Tool for Online Adaptive Radiation Therapy

Amjad, A.; Thill, D.; Lawton, C.A.; O’Connell, Nicolette; Hall, William A.; Li, A.

doi:10.1016/j.ijrobp.2019.06.750

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…For instance, it was shown, that a convolutional neural network can provide a high-quality automatic delineation of the prostate, rectum and bladder with Dice scores of 0.87, 0.89 and 0.95 respectively, but for clinical usage 80% of these automatically made segmentations required manual correction. 6 The common approach is to train one network on all available data without acknowledging the heterogeneous nature of medical data caused by inter-and intra-observer variation. 7 One of the recently introduced approaches aware of observer variation is the probabilistic U-Net.…”

Section: Introduction 11 Backgroundmentioning

confidence: 99%

Observer variation-aware medical image segmentation by combining deep learning and surrogate-assisted genetic algorithms

Dushatskiy

Mendrik

Bosman

et al. 2020

Medical Imaging 2020: Image Processing

View full text Add to dashboard Cite

There has recently been great progress in automatic segmentation of medical images with deep learning algorithms. In most works observer variation is acknowledged to be a problem as it makes training data heterogeneous but so far no attempts have been made to explicitly capture this variation. Here, we propose an approach capable of mimicking different styles of segmentation, which potentially can improve quality and clinical acceptance of automatic segmentation methods. In this work, instead of training one neural network on all available data, we train several neural networks on subgroups of data belonging to different segmentation variations separately. Because a priori it may be unclear what styles of segmentation exist in the data and because different styles do not necessarily map one-on-one to different observers, the subgroups should be automatically determined. We achieve this by searching for the best data partition with a genetic algorithm. Therefore, each network can learn a specific style of segmentation from grouped training data. We provide proof of principle results for open-sourced prostate segmentation MRI data with simulated observer variations. Our approach provides an improvement of up to 23% (depending on simulated variations) in terms of Dice and surface Dice coefficients compared to one network trained on all data.

show abstract

Section: Introduction 11 Backgroundmentioning

confidence: 99%