Identification of endometrial cancer specific miRNA biomarkers in endometrial fluid collected using saline infusion sonohysterography techniques
Year: 2019
Type: article
Abstract: Objectives: Preoperative diagnosis of adnexal tumors as benign or malignant is a necessary step to optimize patient management.We present a Machine Learning (ML) Algorithm to characterize adnexal tumors as benign or malignant from ultrasound images.Methods: Transvaginal and transabdominal 2D B-mode static ultrasound images of 100 adnexal tumors from the prospective IOTA 7 study in Queen Charlotte's and Chelsea Hospital in London were included in the study.The final histology was used as gold standard for all tumors, and the ROI were cropped by the operator.The collaboratively developed ML algorithm automatically extracted nine quantitative texture-based feature vectors: Gabor Filter; Fractal Dimension (FD); Histogram of Gradient; Local Binary Pattern (LBP-256 Bins); 7-moments; Uniform LBP (ULBP-59); Statistics Histogram; skewness and kurtosis.These features are of different dimensions representing the cropped tumor.The Support Vector Machine (SVM) classifier version 1.2 was used to train and test the discriminating diagnostic power of these features.The software outputs the decision of each of the nine models and that of the majority voting fused scheme.Results: By testing 100 adnexal tumors (44 malignant) the performances varied with fusion configurations.Table 1 shows the majority rule results for several odd number of the singular decisions.
DOI: 10.1002/ajum.12144
Language: English
Topic: AI in cancer detection
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W2937907759