2019 · 50 citations · 13 references
ComputingComputer ScienceRadiologyComputer VisionDiagnostic ImagingMachine LearningBreast ImagingCancer DetectionDeep LearningImage PreprocessingMedical Image AnalysisMedical ImagingAdaptive Histogram EqualizationContrast-limited Adaptive Histogram EqualizationData AugmentationDeep Learning In RadiologyDeep Learning-based DiagnosisHistogram EqualizationIncidence RateWrong Diagnosis
The incidence rate of breast cancer continued to rise in the last few decades. Current screening strategy of breast cancer is based on classic X-ray imaging. The sensitivity and specificity of the diagnosis are largely depend on the experiences of the radiologists, and uncertain diagnosis is quite frequent because of resolution limitations and the concerns of lawsuits arisen from wrong diagnosis or undetected lesions. The convolutional neural network is an effective technique for classification in deep learning model. In this study, we utilized median filter, contrast-limited adaptive histogram equalization, and data augmentation to preprocess over 9,000 mammograms, and trained a classified model by using convolutional neural network. The experiment results demonstrated that the accuracy of model with preprocessed images significantly outperformed the model without preprocessed images.
13