Mean Distance to Agreement (Mda)
where h(A, B), denoted as directed HD, given by (h(h(A,B)={max}_{ain A}{min}_{bin B}Green a-bGreenGreen), where a and b represent any point around structures A and B, and (Green a-bGreen) is the Euclidean distance between a and b. When the contours of the two structures are completely coherent, HD and MDA approach zero. The proposed PASSMID worked well with a DSC/MDA in 2 medium dimensions of 0.94/1.78 mm, 0.93/1.04 mm, 0.93/1.06 mm, 0.93/1.14 mm, 0.92/0.83 mm, 0.84/1.53 mm, 0.86/2.39 mm, 0.81/2.49 mm, 0.72/5.48 mm and 0.70/5.03 mm for the liver, left kidney, right kidney, spleen, aorta, pancreas, stomach, duodenum, small intestine and large intestine. From the third fractions, the accuracy of the contours with PASSMID was significantly improved compared to the strategy at a DIR (P < 0.05). The average acceptable bracket ratio was 13.9%, 17.5%, 60.8%, 70.6% and 71.8%, respectively, for the 5 strategies. With regard to DSC (Fig. 3a, b), both SEGatlas and SEGAI showed high segmentation accuracy in the lower jaw and eye. However, low precision has been observed in organs with small volumes, such as the optic nerve and visual chiasm. Alternatively, HD showed comparable or better results in such organs (Fig. 3c, d), due to differences in the derivation of the two methods. The DSC is calculated from the overlap of the two borders and the HD from the simple distance between the two borders. For small volume limits, the DSC tends to be small, although moving both limits is acceptable. This suggests that not only DSC, but also HD or MDA should be used for evaluation and that the use of a single index should be avoided.
When evaluating limits, especially for small volumes, it is advisable to correctly use several indexes and perform visual checks for evaluation. The relative volume errors of the SEGatlas and seGAI were calculated on the basis of the volumes of the manual delimitations. Supplementary file 3: Table S2 shows the relative volume error for each organ. In optical chiasma, the average error was 72.9% in the atlas-based group and 65.7% in the AI-based group. We also looked at the correlation between each DSC, HD and MDA index and volume. Supplementary File 4: Table S3 shows the Pearson correlation coefficients between volume and each index. Independent classification by |%â| The RF are also presented in Table 2, separated by class. Overall, 21 of the 46 radiomic characteristics were classified as robust and 7 of the 13 GLRLM radiomic characteristics as robust.
14 radiomic traits were reported according to the CCC and the average |%â| RF is classified as robust. 24 radiomic characteristics did not have corresponding stability classifications, considering CCC classifications and the mean |% â| RF compares. In addition, the relative volume errors ΔV in the volume of SEGatlas and SEGAI (Vauto) were calculated using the following equation with respect to the manual delimitation volume (Vmanual). In this study, the mean and standard deviation of the relative error were also calculated. Liu, J. et al. Correlation and Agreement: Overview and Clarification of Competing Concepts and Measures. Shanghai Arch. Psychiatry 28 (2), 115â120 (2016). Longitudinal MRIs were used, which were taken on a 1.5 T LINAC MRI for 10 patients with abdominal cancer.
The proposed PASSMID consists of 2 steps: apply a patient-specific vision pipeline to longitudinal MRIs and fill in all contours of previous sessions/fractions to a new fractional MRI using multiple DIR and combine the resulting contours using simultaneous truth and performance level estimation (STAPLE) to achieve final consensus segmentation. Five contour propagation strategies were compared: the planning of fractional MRI computed tomography by rigid body recording (RDR), the pre-treatment MRI with fractional MRI by RDR and DIR, and the proposal of multi-input DIR/STAPLE without pretreatment and passMID. The cube similarity coefficient (DSC) and the mean distance to be matched (MDA) with the contours of the ground truth were calculated layer by layer to quantify the accuracy of the contour. A quantitative index, defined as the ratio of acceptable disks, was introduced using a DSC criterion > 0.8 and MDA < 2 mm. Radiomic characteristics belonging to the class of morphological characteristics can also be used as a measure of contour accuracy. As shown in Table 2, the average is | %â| The RF for the 4 morphological characteristics is less than 0.1%. This small average absolute percentage difference in radiomic characteristic data for the morphological character class is another indication that the automatically generated contour and the manual outline match well. For D0.1cc, D1cc, D2cc and D5cc between two recorded images (primary image and each secondary image), the mean difference ± total SD in the dose distribution between all structures of the DIR process was 0.012 ±0.008 Gy and 0.043 ±0.013 Gy, resulting in 1.05 ±0.26% and 3.58 ±0.83% respectively for small and large deformations. These results showed that the maximum uncertainty of D2cc for the DIR method was about 3.58%. A higher lin concordance correlation coefficient (CCC) results in robust radiomic characteristics. As shown in Table 2, the radiomic characteristics of the NGTDM class had the highest mean CCC of 0.963, while the radiomic characteristics of the GLRLM class had the lowest average of 0.820.
Based on this, NGTDM was the most robust class of radiomic features, while GLRLM was the least robust class of radiomic features when differences in prostate contours were taken into account. Unlike our study, Rizzetto et al. found that GLRLM was the most robust when taking into account the profiled contours of colorectal liver metastases10. These incongruous results may be due to different sites in the body (prostate versus liver) and/or different contour sizes, but only to the idea that the robustness of radiomic characteristics needs to be assessed, as they may vary depending on the location. Future radiomic studies should take into account the site-specific robustness of radiomic characteristics, as data derived from radiomic characteristics have a different contour dependency, as seen in this and other studies3,4,5,10,11,12. The average duration of the RIR and DIR processes was 20 and 50 seconds, respectively. The average ± SD values of DICE, Jaccard, HD and MDA in the RIR and DIR processes is given in Table 2. There were statistically significant differences between the measurements calculated by DIR compared to RIR (p < 0.001 for all measurements), with DIR having the highest accuracy in matching the contours of the bladder and rectum.
In a study by Reniers et al., the mean contour distance of 2 mm or less was proposed for BT dosimetry [21]. In this study, we received 0.72 mm and 1.19 mm for the bladder and rectum, respectively, which are better results compared to the proposal by Reniers et al. Several previous studies have found no significant differences between the DIR and SS approaches that could be related to patient sample size, number of treatment fractions, or the DIR algorithm used [21,30,32]. We considered values above Q3+1.5 IQR as outliers (Table 4) and found statistically significant differences between the DIR and SS approaches due to the appropriate sample size and a hybrid DIR algorithm with high accuracy for at least 90% of patients. Various measures have been developed to assess the accuracy of DIR [26,27,28]. We used the distance measurements DICE, Jaccard, Hausdorff (HD) and MDA (Mean Distance to Agreement) between the structures to quantitatively calculate the uncertainty of contour adjustment in the virtual phantom. We also calculated these measures for patient data. DICE and Jaccard are given as follows: Table 1 summarizes the mean, standard deviation and range for DSC, MDA, ÎCM, âVol, JD Minimum and JD Maximum between the automatic and manual contours of the prostate, and Fig. 2 shows the distribution of DSC, MDA, ÎCM and ÎVol over 1010 fractions. The average DSC of all 1,010 fractions, 0.90 ±0.04, is within the DSC value of â~0.8 to 0.9 recommended by TG-132. A total of 42 fractions had DSCs< 0.8, which is below the lower tolerance recommendation of TG-132. The quantitative evaluation of the recorded prostate contours was carried out on the basis of several measurements. The cube similarity coefficient (DSC) is a statistical measure of contour overlap, where 0 is not an overlap and 1 is a perfect match.27 The tuning distance (DTA) between two edges, sometimes referred to as the distance to compliance, is the shortest distance between a particular point on the surface of one stroke and the surface of the other stroke.
The mean distance from the chord (MDA) is the average of all DTA28 distances. The geometric centers of the automatic and manual contours on the CBCTs were calculated and used to determine the difference in the center of gravity of the center of mass (ÎCM). For the 1,010 fractions, the volume difference between the automatic and manual contour (âVol) was also evaluated. For a smaller subsample of 10 patients, the Jacobin determinant (JD) was calculated every two weeks (a total of 47 fractions). The Jacobic defining values corresponding to the expansion of the volume are not respectively the volume variation and the volume reduction>1, 1 and <1. JD values equal to or less than zero correspond to non-physical transformations that indicate a bad DIR20. DSC, MDA, ÎCM and âVol were calculated between the automatic and manual contours of the prostate for the 1,010 fractions. .
最新記事 by kabumori@yamanouchi (全て見る)
- 5.8 Agreement of Subject and Verb Latin - 2022年6月8日
- Y(Uk)3 the Contracts (Rights of Third Parties) Act 1999 - 2022年4月21日
- Work Agreement Duration - 2022年4月20日
