Sensitivity of deep learning applied to spatial image steganalysis.

In recent years, the traditional approach to spatial image steganalysis has shifted to deep learning (DL) techniques, which have improved the detection accuracy while combining feature extraction and classification in a single model, usually a convolutional neural network (CNN). The main contribution from researchers in this area is new architectures that further improve detection accuracy. Nevertheless, the preprocessing and partition of the database influence the overall performance of the CNN. This paper presents the results achieved by novel steganalysis networks (Xu-Net, Ye-Net, Yedroudj-Net, SR-Net, Zhu-Net, and GBRAS-Net) using different combinations of image and filter normalization ranges, various database splits, different activation functions for the preprocessing stage, as well as an analysis on the activation maps and how to report accuracy. These results demonstrate how sensible steganalysis systems are to changes in any stage of the process, and how important it is for researchers in this field to register and report their work thoroughly. We also propose a set of recommendations for the design of experiments in steganalysis with DL.

Strategy to improve the accuracy of convolutional neural network architectures applied to digital image steganalysis in the spatial domain.

In recent years, Deep Learning techniques applied to steganalysis have surpassed the traditional two-stage approach by unifying feature extraction and classification in a single model, the Convolutional Neural Network (CNN). Several CNN architectures have been proposed to solve this task, improving steganographic images' detection accuracy, but it is unclear which computational elements are relevant. Here we present a strategy to improve accuracy, convergence, and stability during training. The strategy involves a preprocessing stage with Spatial Rich Models filters, Spatial Dropout, Absolute Value layer, and Batch Normalization. Using the strategy improves the performance of three steganalysis CNNs and two image classification CNNs by enhancing the accuracy from 2% up to 10% while reducing the training time to less than 6 h and improving the networks' stability.

Representation learning for mammography mass lesion classification with convolutional neural networks.

BACKGROUND AND OBJECTIVE: The automatic classification of breast imaging lesions is currently an unsolved problem. This paper describes an innovative representation learning framework for breast cancer diagnosis in mammography that integrates deep learning techniques to automatically learn discriminative features avoiding the design of specific hand-crafted image-based feature detectors. METHODS: A new biopsy proven benchmarking dataset was built from 344 breast cancer patients' cases containing a total of 736 film mammography (mediolateral oblique and craniocaudal) views, representative of manually segmented lesions associated with masses: 426 benign lesions and 310 malignant lesions. The developed method comprises two main stages: (i) preprocessing to enhance image details and (ii) supervised training for learning both the features and the breast imaging lesions classifier. In contrast to previous works, we adopt a hybrid approach where convolutional neural networks are used to learn the representation in a supervised way instead of designing particular descriptors to explain the content of mammography images. RESULTS: Experimental results using the developed benchmarking breast cancer dataset demonstrated that our method exhibits significant improved performance when compared to state-of-the-art image descriptors, such as histogram of oriented gradients (HOG) and histogram of the gradient divergence (HGD), increasing the performance from 0.787 to 0.822 in terms of the area under the ROC curve (AUC). Interestingly, this model also outperforms a set of hand-crafted features that take advantage of additional information from segmentation by the radiologist. Finally, the combination of both representations, learned and hand-crafted, resulted in the best descriptor for mass lesion classification, obtaining 0.826 in the AUC score. CONCLUSIONS: A novel deep learning based framework to automatically address classification of breast mass lesions in mammography was developed.

Convolutional neural networks for mammography mass lesion classification.

Feature extraction is a fundamental step when mammography image analysis is addressed using learning based approaches. Traditionally, problem dependent handcrafted features are used to represent the content of images. An alternative approach successfully applied in other domains is the use of neural networks to automatically discover good features. This work presents an evaluation of convolutional neural networks to learn features for mammography mass lesions before feeding them to a classification stage. Experimental results showed that this approach is a suitable strategy outperforming the state-of-the-art representation from 79.9% to 86% in terms of area under the ROC curve.

Discovering mammography-based machine learning classifiers for breast cancer diagnosis.

This work explores the design of mammography-based machine learning classifiers (MLC) and proposes a new method to build MLC for breast cancer diagnosis. We massively evaluated MLC configurations to classify features vectors extracted from segmented regions (pathological lesion or normal tissue) on craniocaudal (CC) and/or mediolateral oblique (MLO) mammography image views, providing BI-RADS diagnosis. Previously, appropriate combinations of image processing and normalization techniques were applied to reduce image artifacts and increase mammograms details. The method can be used under different data acquisition circumstances and exploits computer clusters to select well performing MLC configurations. We evaluated 286 cases extracted from the repository owned by HSJ-FMUP, where specialized radiologists segmented regions on CC and/or MLO images (biopsies provided the golden standard). Around 20,000 MLC configurations were evaluated, obtaining classifiers achieving an area under the ROC curve of 0.996 when combining features vectors extracted from CC and MLO views of the same case.

A software framework for building biomedical machine learning classifiers through grid computing resources.

This paper describes the BiomedTK software framework, created to perform massive explorations of machine learning classifiers configurations for biomedical data analysis over distributed Grid computing resources. BiomedTK integrates ROC analysis throughout the complete classifier construction process and enables explorations of large parameter sweeps for training third party classifiers such as artificial neural networks and support vector machines, offering the capability to harness the vast amount of computing power serviced by Grid infrastructures. In addition, it includes classifiers modified by the authors for ROC optimization and functionality to build ensemble classifiers and manipulate datasets (import/export, extract and transform data, etc.). BiomedTK was experimentally validated by training thousands of classifier configurations for representative biomedical UCI datasets reaching in little time classification levels comparable to those reported in existing literature. The comprehensive method herewith presented represents an improvement to biomedical data analysis in both methodology and potential reach of machine learning based experimentation.

Grid infrastructures for developing mammography CAD systems.

This paper presents a set of technologies developed to exploit Grid infrastructures for breast cancer CAD, that include (1) federated repositories of mammography images and clinical data over Grid storage, (2) a workstation for mammography image analysis and diagnosis and (3) a framework for data analysis and training machine learning classifiers over Grid computing power specially tuned for medical image based data. An experimental mammography digital repository of approximately 300 mammograms from the MIAS database was created and classifiers were built achieving a 0.85 average area under the ROC curve in a dataset of 100 selected mammograms with representative pathological lesions and normal cases. Similar results were achieved with classifiers built for the UCI Breast Cancer Wisconsin dataset (699 features vectors). Now these technologies are being validated in a real medical environment at the Faculty of Medicine in Porto University after a process of integrating the tools within the clinicians workflows and IT systems.