Mathematical modelling to compute ground truth from 3D images is an area of research that can strongly benefit from machine learning methods. Deep neural networks (DNNs) are state-of-the-art methods design for solving these kinds of difficulties. Convolutional neural networks (CNNs), as one class of DNNs, can overcome special requirements of quantitative analysis especially when image segmentation is needed. This article presents a system that uses a cascade of CNNs with symmetric blocks of layers in chain, dedicated to 3D image segmentation from microscopic images of 3D nuclei. The system is designed through eight experiments that differ in following aspects: number of training slices and 3D samples for training, usage of pre-trained CNNs and number of slices and 3D samples for validation. CNNs parameters are optimized using linear, brute force, and random combinatorics, followed by voter and median operations. Data augmentation techniques such as reflection, translation and rotation are used in order to produce sufficient training set for CNNs. Optimal CNN parameters are reached by defining 11 standard and two proposed metrics. Finally, benchmarking demonstrates that CNNs improve segmentation accuracy, reliability and increased annotation accuracy, confirming the relevance of CNNs to generate high-throughput mathematical ground truth 3D images.

Digital analysis and biomedical image processing has become important part within modern medicine and biology. Digital pathology is just one of many medicine areas that is being upgraded by constant biomedical engineering research and development. It is very important that some of disciplines as nucleus detection, image segmentation or classification become more and more effective, with minimum human intervention on these processes, and maximum accuracy and precision. Improved optimization of nucleus segmentation methods parameters based on two levels of voting processes is presented in this paper. First level includes hybrid nucleus segmentation based on 7 segmentation algorithms: OTSU, Adaptive Fuzzy-c means, Adaptive K-means, KGB (Kernel Graph Cut), APC (Affinity Propagation Clustering), Multi Modal and SRM (Statistical region merging) based on optimization of algorithms parameters along with implemented first level voting structure. Second level voting structure includes segmentation results obtained in the first level of voting structure in combination with 3rd party segmentation tools: ImageJ/Fiji and MIB (Microscopy Image Browser). A definite segmented image of a nucleus could serve as a generic ground truth image because it is formed as a result of a consensus based on several different methods of segmentation and different parameter settings, which guarantees better objectivity of the results. In addition, this approach can be used with great scalability on 3D-stack image datasets.

With rising cancer rates in world, it is important to incorporate all possible ways in order to prevent, detect, and cure this disease. Breast cancer presents one of those threats, and bioinformatics field must work towards finding models to fight against it, with one of them being creation of classification model for that kind of illness. Using machine learning techniques in order to make these classifications is one of those ways. It is widely known that ANN (Artificial Neural Network) and ANFIS (Adaptive Neurofuzzy Inherent System) can significantly upgrade any kind of classification process, and in that way, help in biomedicine and cancer treatment. Furthermore, more objective models for classification must be developed, regarding both time and resources, in order to get optimal results. In this paper, GA (Genetic Algorithm) algorithm that optimize ANN and ANFIS has been used to make classification of breast cancer diagnosis. It is shown that GA optimization of ANFIS and ANN parameters results in creating model with better accuracy comparing to basic classifiers. Voting method has been used on such GA ANFIS optimized structure, in order to achieve model with higher reliability. Final score of computed models was determined using external validation, based on 4 most relevant clinical metrics: sensitivity, specificity, accuracy and precision.

This paper presents the results of the analysis of the network intrusion detection systems using data mining techniques and anomaly detection. Anomaly detection technique is present for a while in the area of data mining. Previous papers that implement data mining techniques to detect anomaly attacks actually use well-known techniques such as classification or clustering. Anomaly detection technique combines all these techniques. They are also facing problem on the fact that many of the attacks do not have some kind of signature on network and transport layer, so it is not easy to train models for these type of attacks. Network dataset that was used in this paper is DARPA 1998 dataset created in MIT Lincoln Laboratory and is used worldwide for the network testing purposes.