ABSTRACT
OBJECTIVE@#To explore the feasibility of using the bidirectional local distance based medical similarity index (MSI) to evaluate automatic segmentation on medical images.@*METHODS@#Taking the intermediate risk clinical target volume for nasopharyngeal carcinoma manually segmented by an experience radiation oncologist as region of interest, using Atlas-based and deep-learning-based methods to obtain automatic segmentation respectively, and calculated multiple MSI and Dice similarity coefficient (DSC) between manual segmentation and automatic segmentation. Then the difference between MSI and DSC was comparatively analyzed.@*RESULTS@#DSC values for Atlas-based and deep-learning-based automatic segmentation were 0.73 and 0.84 respectively. MSI values for them varied between 0.29~0.78 and 0.44~0.91 under different inside-outside-level.@*CONCLUSIONS@#It is feasible to use MSI to evaluate the results of automatic segmentation. By setting the penalty coefficient, it can reflect phenomena such as under-delineation and over-delineation, and improve the sensitivity of medical image contour similarity evaluation.
Subject(s)
Feasibility Studies , Radiotherapy Planning, Computer-AssistedABSTRACT
Objective To perform a preclinical test of a delineation software based on atlas-based auto-segmentation (ABAS),to evaluate its accuracy in the delineation of organs at risk (OARs) in radiotherapy planning for nasopharyngeal carcinoma (NPC),and to provide a basis for its clinical application.Methods Using OARs manually contoured by physicians on planning-CT images of 22 patients with NPC as the standard,the automatic delineation using two different algorithms (general and head/neck) of the ABAS software were applied to the following tests:(1) to evaluate the restoration of the atlas by the software,automatic delineation was performed on copied images from each patient using the contours of OARs manually delineated on the original images as atlases;(2) to evaluate the accuracy of automatic delineation on images from various patients using a single atlas,the contours manually delineated on images from one patients were used as atlases for automatic delineation of OARs on images from other patients.Dice similarity coefficient (DSC),volume difference (Vdiff),correlation between the DSC and the volume of OARs,and efficiency difference between manual delineation and automatic delineation plus manual modification were used as indices for evaluation.Wilcoxon signed rank test and Spearman correlation analysis were used.Results The head/neck algorithm had superior restoration of the atlas over the general algorithm.The DSC was positively correlated with the volume of OARs and was higher than 0.8 for OARs larger than 1 cc in volume in the restoration test.For automatic delineation with the head/neck algorithm using a single atlas,the mean DSC and Vdiff were 0.81-0.90 and 2.73％-16.02％,respectively,for the brain stem,temporal lobes,parotids,and mandible,while the mean DSC was 0.45-0.49 for the temporomandibular joint and optic chiasm.Compared with manual delineation,automatic delineation plus manual modification saved 68％ of the time.Conclusions A preclinical test is able to determine the accuracy and conditions of the ABAS software in specific clinical application.The tested software can help to improve the efficiency of OAR delineation in radiotherapy planning for NPC.However,it is not suitable for delineation of OAR with a relatively small volume.