2018 · 25 citations · 12 references
ComputingComputer ScienceRadiologyComputer VisionDiagnostic ImagingMachine LearningCancer DetectionCancer DiagnosisDeep LearningImage PreprocessingMedical Image AnalysisMedical ImagingConvolutionConvolutional Neural NetworksDeep Learning-based ClassificationDeep Learning-based DiagnosisEarly DiagnosisGlobal Contrast EnhancementModel Performance AnalysisNormalization
Computerized breast cancer diagnosis system has played an import role in early cancer diagnosis. For this purpose, we apply deep learning by using convolutional neural networks (CNN) to classify abnormalities, benign or malignant, in mammographic images based on the mini Mammographic Image Analysis Society (mini-MIAS) database. Accuracy, sensitivity, and specific values are observed to evaluate the performance of the CNN. To improve the performance, we utilize image-preprocessing methods containing cropping, global contrast normalization, augmentation, local histogram equalization, and balancing preprocessing. We built four CNN models to study the impact of depth and hidden layer structure on model performance. The CNN-4d model performs best among four proposed CNN models consisting of four convolution layers with a dropout of 0.7. The CNN-4d model achieved a balance of high sensitivity (90.63%) and high specificity (87.67%), and an accuracy of 89.05%. The result of this study indicates that CNNs have promising potential in the field of intelligent medical image diagnosis.
12
Dropout: a simple way to prevent neural networks from overfitting
Nitish Srivastava, Geoffrey E. Hinton, Alex Krizhevsky et al. · 2014 · 34.3K citations
On the importance of initialization and momentum in deep learning
Ilya Sutskever, James Martens, George E. Dahl et al. · 2013 · 3.5K citations
What is the best multi-stage architecture for object recognition?
Kevin Jarrett, Koray Kavukcuoglu, M. Ranzato et al. · 2009 · 2.2K citations
Flexible, high performance convolutional neural networks for image classification
Dan Cireşan, Ueli Meier, Jonathan Masci et al. · 2011 · 1.2K citations