RESUMO
Objective To evaluate the autocontouring accuracy using the atlas-based autosegmentation of CT images for head-and-neck cancer.Methods Ten head and neck patients with contours were selected.Two groups of autocontouring atlas were tested,the first group was for patients with own atlas,for the second group we tested the autocontouring of eight patients with other two patients atlas.Dice similarity coefficient (DSC) and overlap index (OI) were introduced to evaluate the autocontours,and the discrepancy between the two groups was evaluated through paired t-test.Results Both the DSC and OIof all the organs in the first group were ＞0.80,the result of mandible was the highest ( ＞0.91 ),the DSC of the gross tumor volume (GTV) was the lowest (0.81 ),the OI of the GTV was 0.82,and the DSC and OI of the clinical target volume (node) were 0.82 and 0,79,respectively.Only the risk organ was delineated in the second group,and spinal cord and brain stem were combined to analyze.All the DSC was about 0.70,and the DSC and OI of mandible were higher than the others,which was due to its bone anatomy.The accuracy in the second group was significantly lower than that of the first group ( t =3.24 - 8.26,P =0.014 -0.000),except the right parotid (t=2.08,P=0.075).Conclusions Automatic segmentation generates contours of sufficient accuracy for adaptive planning intensity-modulated radiotherapy (IMRT) to accommodate anatomic changes during treatment.For convention planning IMRT normal structure auto-contouring,it need to select more standard atlas in order to acquire a satisfied autocontours.
