Discovery of adversarial attacks on deep neural networks, have exposed the vulnerabilities of these networks, wherein they often entirely fail to classify the attack generated images. While deep learning networks have generated promising results in performing brain tumor classification, there has been no analysis of their susceptibility to adversarial attacks. Vulnerability to adversarial attacks can render the deep neural networks useless for practical medical application. In this paper, a study has been performed to determine extent of white box adversarial attacks on convolutional neural networks used for brain tumor classification. Three different adversarial attacks are implemented on the network, namely Noise generated, Fast Gradient Sign, and Virtual Adversarial Training methods. The performance of the network under these attacks is analyzed and discussed. Furthermore, in the paper it is shown how these networks perform when trained on the adversarial attack generated images, which could be a possible solution to prevent the failure of the classification networks against adversarial attacks.
View publication
View paper

As cancer treatments are gaining momentum, in a bid to improve drug potency, doctors are looking towards precision cancer medicine. Here the drug prescriptions are tailored to the patients gene changes. In this paper, we aim to automate the task of drug selection, by predicting the clinical outcome of using a particular drug on a combination of the patients gene, variant and cancer disease type. While the main idea behind precision cancer treatment is to identify drugs suitable to each patients unique case, it is justifiable for us to assume that there exists a predictive pattern in these prescriptions. We propose to implement this prediction using three machine learning models, the Support Vector Machine, the Random Forest Classifier and the Deep Neural Network. The models yielded promising results of over 90% accuracy and over 95% ROC-AUC score. This positive outcome affirms our assumption that there exists a predictive pattern in precision treatments, that could be extrapolated to help automate such tasks. We further analyzed the data set and identified latent relations between drug, cancer disease, target gene and gene variant. This exploration uncovered some significant patterns where we can observe how a particular drug has had successful results in treating a particular cancer and targeting specific gene variants.
View publication
View paper

Face Recognition is steadily making its way into commercial products. As such, the accuracy of Face Recognition systems is becoming extremely crucial. In the Firefly Algorithm, the brightness of fireflies is used to measure attraction between a pair of unisex fireflies. The firefly with higher brightness attracts the less bright firefly. The objective function is defined in proportion to the brightness, to define a maximization problem. This chapter aims to present the promising application of the Firefly Algorithm for Face Recognition. The Firefly Algorithm is used in a hyper dimensional feature space to select features that maximize the recognition accuracy. This chapter delineates how the Firefly Algorithm is a suitable algorithm for selection of the features in a Face Recognition model. The Firefly Algorithm is then applied on this feature space to identify and select the best features. Fireflies are arbitrarily placed on various focal points of the image under consideration. The advantage of this approach is its fast convergence in selecting the best features. The gamma parameter (γ) controls the movement of fireflies in this feature space and can be tuned for gaining an improvement in the performance of the Face Recognition model. This chapter aims to evaluate the performance and viability of using the Firefly Algorithm for Face Recognition.
View chapter

Cuckoo Search (CS) is a nature inspired and meta-heuristic algorithm which is based on a brood reproductive strategy of cuckoo birds to increase their population. This algorithm mainly serves to determine the maximum or minimum value of a particular problem which is known as the objective function. CS has reportedly outperformed other nature inspired algorithms in terms of computational efficiency and the speed of convergence to reach an optimal solution. This chapter aims at exploring the application of CS to determine the parameters of Artificial Neural Networks (ANN). The inherent problem with traditional training of ANNs using backpropagation is that the learning process cannot guarantee a global minimum solution and has a tendency of getting trapped in a local minima. The working of such ANN models are restricted to a differentiable neuron transfer function. The CS algorithm has been observed to provide a solution without the use of derivates to optimize such convoluted problems. The usage of ANNs across a wide range of problems including classification tasks, image processing, signal processing, etc. justifies the application of CS to the backpropagation procedure of ANNs to achieve a faster rate of convergence and avoid the local minima problem. This chapter also presents discussions and results on how ANNs optimized with variants of CS perform when applied to detection of chronic kidney disease, modeling of operating photovoltaic module temperature and forest type classification.
View chapter

While deep learning systems have provided breakthroughs in several tasks in the medical domain, they are still limited by a problem faced by most deep learning applications. That problem is the dependency on the availability of training data. To counter this limitation, there is active research ongoing in few shot learning. Few shot learning algorithms aim to overcome the data dependency by exploiting the information available from a very small amount of data. The success of these algorithms can prove to be a significant advancement in the application of deep learning systems. In medical imaging, there is often a limitation on the available data, due to the rare occurrence of some diseases. As a result, there are several potential use cases for few shot learning algorithms in medical imaging. In this chapter, the background and working of few shot learning algorithms is explained. Then there is a study of the problems faced in medical imaging related to availability of limited data. After establishing context, the recent advances the in application of few shot learning to medical imaging tasks such as classification and segmentation are explored. The results of these applications are examined with a discussion on its future scope.
View chapter