Yuta Akamine1, Yoshiko Ueno2, Keitaro Sofue2, Takamichi Murakami2, Yu Ueda1, Ahsan Budrul1, Masami Yoneyama1, Makoto Obara1, and Marc Van Cauteren3
1Philips Japan, Tokyo, Japan, 2Department of Radiology, Kobe University Graduate School of Medicine, Hyogo, Japan, 3Asia Pacific, Philips Healthcare, Tokyo, Japan
1Philips Japan, Tokyo, Japan, 2Department of Radiology, Kobe University Graduate School of Medicine, Hyogo, Japan, 3Asia Pacific, Philips Healthcare, Tokyo, Japan
SMOTE as data augmentation and ensemble learning are combined, trained
from mp-MR including IVIM, DKI, and permeability. This proposed method showed F1 (0.831) and AUC (0.762) and is
effective strategy to improve diagnosis of low from high grade prostate cancer for
imbalanced small dataset.
Fig. 1. Schema of synthetic-minority-over-sampling-technique (SMOTE) algorism. (A) SMOTE algorithm generates synthetic examples
through a linear interpolation between two existing minority examples. (B)
For each minority sample, depending on the amount of synthetic
examples, neighbors from the k nearest neighbors are randomly chosen. In this
study, we used five nearest neighbors.
Table 4. Performance of proposed method
combining SMOTE and ensemble model, compared to base models. Accuracy,
sensitivity, specificity, F1 score, and AUC are evaluated. P value between
ensemble model without SMOTE and with
SMOTE is shown. A paired t-test was used and a P- value less than 0.05 was
considered significant.
