Are deep learning classification results obtained on CT scans fair and interpretable?

Following the great success of various deep learning methods in image and object classification, the biomedical image processing society is also overwhelmed with their applications to various automatic diagnosis cases. Unfortunately, most of the deep learning-based classification attempts in the literature solely focus on the aim of extreme accuracy scores, without considering interpretability, or patient-wise separation of training and test data. For example, most lung nodule classification papers using deep learning randomly shuffle data and split it into training, validation, and test sets, causing certain images from the Computed Tomography (CT) scan of a person to be in the training set, while other images of the same person to be in the validation or testing image sets. This can result in reporting misleading accuracy rates and the learning of irrelevant features, ultimately reducing the real-life usability of these models. When the deep neural networks trained on the traditional, unfair data shuffling method are challenged with new patient images, it is observed that the trained models perform poorly. In contrast, deep neural networks trained with strict patient-level separation maintain their accuracy rates even when new patient images are tested. Heat map visualizations of the activations of the deep neural networks trained with strict patient-level separation indicate a higher degree of focus on the relevant nodules. We argue that the research question posed in the title has a positive answer only if the deep neural networks are trained with images of patients that are strictly isolated from the validation and testing patient sets.

A New Feature Ensemble with a Multistage Classification Scheme for Breast Cancer Diagnosis.

A new and effective feature ensemble with a multistage classification is proposed to be implemented in a computer-aided diagnosis (CAD) system for breast cancer diagnosis. A publicly available mammogram image dataset collected during the Image Retrieval in Medical Applications (IRMA) project is utilized to verify the suggested feature ensemble and multistage classification. In achieving the CAD system, feature extraction is performed on the mammogram region of interest (ROI) images which are preprocessed by applying a histogram equalization followed by a nonlocal means filtering. The proposed feature ensemble is formed by concatenating the local configuration pattern-based, statistical, and frequency domain features. The classification process of these features is implemented in three cases: a one-stage study, a two-stage study, and a three-stage study. Eight well-known classifiers are used in all cases of this multistage classification scheme. Additionally, the results of the classifiers that provide the top three performances are combined via a majority voting technique to improve the recognition accuracy on both two- and three-stage studies. A maximum of 85.47%, 88.79%, and 93.52% classification accuracies are attained by the one-, two-, and three-stage studies, respectively. The proposed multistage classification scheme is more effective than the single-stage classification for breast cancer diagnosis.

A new classification scheme of plastic wastes based upon recycling labels.

Since recycling of materials is widely assumed to be environmentally and economically beneficial, reliable sorting and processing of waste packaging materials such as plastics is very important for recycling with high efficiency. An automated system that can quickly categorize these materials is certainly needed for obtaining maximum classification while maintaining high throughput. In this paper, first of all, the photographs of the plastic bottles have been taken and several preprocessing steps were carried out. The first preprocessing step is to extract the plastic area of a bottle from the background. Then, the morphological image operations are implemented. These operations are edge detection, noise removal, hole removing, image enhancement, and image segmentation. These morphological operations can be generally defined in terms of the combinations of erosion and dilation. The effect of bottle color as well as label are eliminated using these operations. Secondly, the pixel-wise intensity values of the plastic bottle images have been used together with the most popular subspace and statistical feature extraction methods to construct the feature vectors in this study. Only three types of plastics are considered due to higher existence ratio of them than the other plastic types in the world. The decision mechanism consists of five different feature extraction methods including as Principal Component Analysis (PCA), Kernel PCA (KPCA), Fisher's Linear Discriminant Analysis (FLDA), Singular Value Decomposition (SVD) and Laplacian Eigenmaps (LEMAP) and uses a simple experimental setup with a camera and homogenous backlighting. Due to the giving global solution for a classification problem, Support Vector Machine (SVM) is selected to achieve the classification task and majority voting technique is used as the decision mechanism. This technique equally weights each classification result and assigns the given plastic object to the class that the most classification results agree on. The proposed classification scheme provides high accuracy rate, and also it is able to run in real-time applications. It can automatically classify the plastic bottle types with approximately 90% recognition accuracy. Besides this, the proposed methodology yields approximately 96% classification rate for the separation of PET or non-PET plastic types. It also gives 92% accuracy for the categorization of non-PET plastic types into HPDE or PP.

A new feature extraction framework based on wavelets for breast cancer diagnosis.

This paper investigates a pattern recognition framework in order to determine and classify breast cancer cases. Initially, a two-class separation study classifying normal and abnormal (cancerous) breast tissues is achieved. The Histogram of Oriented Gradients (HOG), Dense Scale Invariant Feature Transform (DSIFT), and Local Configuration Pattern (LCP) methods are used to extract the rotation- and scale-invariant features for all tissue patches. A classification is made utilizing Support Vector Machine (SVM), k-Nearest Neighborhood (k-NN), Decision Tree, and Fisher Linear Discriminant Analysis (FLDA) via 10-fold cross validation. Then, a three-class study (normal, benign, and malignant cancerous cases) is carried out using similar procedures in a two-class case; however, the attained classification accuracies are not sufficiently satisfied. Therefore, a new feature extraction framework is proposed. The feature vectors are again extracted with this new framework, and more satisfactory results are obtained. Our new framework achieved a remarkable increase in recognition performance for the three-class study.