China
Africa
Explore chapters and articles related to this topic
Machine learning for radiation oncology
Published in Jun Deng, Lei Xing, Big Data in Radiation Oncology, 2019
Statistical tests come in two forms: parametric and nonparametric, where the former make strong assumptions about the distribution of the underlying data, and the latter make weaker assumptions about the data but are also typically less powerful than the former. Regarding the type of problem, the t-test (parametric), McNemar’s test (nonparametric), and sign test are frequently used for the comparison of two algorithms on a single domain; the sign test (nonparametric), and Wilcoxon’s signed-rank test (nonparametric) are designed for the comparison of two algorithms on several domains; the Friedman’s test (nonparametric) and Nemenyi test are employed for the comparison of multiple algorithms over multiple domains. Although it is often difficult, if not impossible, to verify that all the assumptions hold in these statistical tests, and the results of statistical tests are often misinterpreted, it is always possible to show that a difference between two alternatives, no matter how small, is significant, provided that enough data are used. However, machine learning and data mining researchers should know the applicability and limitations of statistical methods and decide on their own when a statistical test is warranted and when the search for new ideas may be necessary. [33]
Clustering-based undersampling with random over sampling examples and support vector machine for imbalanced classification of breast cancer diagnosis
Published in Computer Assisted Surgery, 2019
In order to evaluate the significance of results from the comparative methods, Friedman test with 95% confidence level [26] is carried out on six datasets. All the methods on six datasets are sorted according to the mean ranks on their MCC performance measures, since as the previous analysis if the data are unbalanced, computing MCC of classification system can be much more appropriate than computing other measures [27]. The null alternative hypothesis is that there is no significant difference between the methods on six datasets. Subsequently, Table 7 displays the p-value which is less than 0.05, the test result rejects the null hypothesis, i.e., there are some differences between these methods. After that, we performed a post hot test to identify if the methods have significant difference [28]. A Nemenyi test at the significance level
In Vitro Biomechanical Study of Epidural Pressure during the Z-shape Elevating-Pulling Reduction Technique for Cervical Unilateral Locked Facets
Published in Journal of Investigative Surgery, 2019
Xinwei Shao, Jican Zeng, Yuchun Chen, Lixian Wu, Xinjia Wang
Statistical analyses of data were performed using SPSS v20.0 (SPSS, Chicago, IL, USA). The “K related samples tests” and “two related samples tests” were used to analyze the non-normally distributed data, followed by the Nemenyi test. K related samples tests were carried out to determine significant differences in the general comparison of epidural pressures in all positions, with P < 0.05 considered significant. Two related samples tests were carried out to determine significant differences in the anterior and posterior epidural pressures between each position; P < 0.005 was considered significant, as the number of comparison times was 10, and α was adjusted to 0.005.
MRI-based automatic segmentation of rectal cancer using 2D U-Net on two independent cohorts
Published in Acta Oncologica, 2022
Franziska Knuth, Ingvild Askim Adde, Bao Ngoc Huynh, Aurora Rosvoll Groendahl, René Mario Winter, Anne Negård, Stein Harald Holmedal, Sebastian Meltzer, Anne Hansen Ree, Kjersti Flatmark, Svein Dueland, Knut Håkon Hole, Therese Seierstad, Kathrine Røe Redalen, Cecilia Marie Futsaether
Figure 2(A) summarizes the resulting DSCp of the five-fold cross-validation on C1, where learning rate, loss function and input normalization were varied. The Friedman test detected an effect of training parameter variation (p < 0.0001) for models based on T2w or T2w + DW images. Thus, a post hoc Nemenyi test was used, and the results are shown in Figure 2(B).