Segmentation precision of abdominal anatomy for MRI-based radiotherapy

Abstract

The limited soft tissue visualization provided by computed tomography, the standard imaging modality for radiotherapy treatment planning and daily localization, has motivated studies on the use of magnetic resonance imaging (MRI) for better characterization of treatment sites, such as the prostate and head and neck. However, no studies have been conducted on MRI-based segmentation for the abdomen, a site that could greatly benefit from enhanced soft tissue targeting. We investigated the interobserver and intraobserver precision in segmentation of abdominal organs on MR images for treatment planning and localization. Manual segmentation of 8 abdominal organs was performed by 3 independent observers on MR images acquired from 14 healthy subjects. Observers repeated segmentation 4 separate times for each image set. Interobserver and intraobserver contouring precision was assessed by computing 3-dimensional overlap (Dice coefficient [DC]) and distance to agreement (Hausdorff distance [HD]) of segmented organs. The mean and standard deviation of intraobserver and interobserver DC and HD values were DCintraobserver = 0.89 ± 0.12, HDintraobserver = 3.6 mm ± 1.5, DCinterobserver = 0.89 ± 0.15, and HDinterobserver = 3.2 mm ± 1.4. Overall, metrics indicated good interobserver/intraobserver precision (mean DC > 0.7, mean HD < 4 mm). Results suggest that MRI offers good segmentation precision for abdominal sites. These findings support the utility of MRI for abdominal planning and localization, as emerging MRI technologies, techniques, and onboard imaging devices are beginning to enable MRI-based radiotherapy.

A novel MRI segmentation method using CNN‐based correction network for MRI‐guided adaptive radiotherapy

“…The average Dice coefficients were calculated and reported in Table for the 10 testing datasets. The CNN+CRF method could remove some of the spurious labels and make the results more plausible than the CNN.…”

“…Accurate contouring of critical OARs is one of the keys to MR‐IGART. Contouring uncertainties can translate into undesired dose distributions on actual anatomy; however, manual OAR contouring is a time‐consuming process which is subject to inter‐observer variation . The time required by the contouring process is a principal bottleneck for treatment plan adaptation.…”

An adaptive Fuzzy C-means method utilizing neighboring information for breast tumor segmentation in ultrasound images

“…Besides, the Hausdorff distance measures the similarity between the neighbors, which incorporates additional information in distinguishing pixels. Actually, the definition measuring the similarity between the two sets was used to evaluate segmentation results . This improved standard Euclid distance based segmentation, which may also be potentially useful for segmenting images from other modalities …”

“…Actually, the definition measuring the similarity between the two sets was used to Medical Physics, 44 (7), July 2017 evaluate segmentation results. 40,41 This improved standard Euclid distance based segmentation, 42 which may also be potentially useful for segmenting images from other modalities. [43][44][45] Regarding the FCM based segmentations of the breast tumors, most studies were aimed for MR images, for example, FCM based adaptive k-means cluster to segment breast tissue MR images.…”

A feasibility study evaluating the relationship between dose and focal liver reaction in stereotactic ablative radiotherapy for liver cancer based on intensity change of Gd-EOB-DTPA-enhanced magnetic resonance images

“…However, as the signal intensity of Gd-EOB-DTPA-enhanced liver MRI was not calibrated or quantified for hepatic function, the volume of FLR was identified by a physician on Gd-EOB-DTPA-enhanced MRI in the studies, and TD was obtained with the DVH analysis [ 21 22 ]. Although use of MRI could increase accuracy of target delineation [ 23 ], inter- and intraobserver variation still exist [ 24 25 26 ]. In the methods of the DVH analysis, as the volume of FLR was determined by the physicians, inter- and intraobserver variation could affect the determination of the TD.…”