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.

The aim of this research is to automate an analysis of the EGFR gene as a whole, and especially an analysis of those exons with clinically identified microdeletion mutations which are recorded with non-mutated nucleotides in a long chains of a, c, t, g nucleotides, and “-” (microdeletion) in the NCBI database or other sites. In addition, the developed system can analyze data resulting from EGFR gene DNA sequencing or DNA extraction for a new patient and identify regions potential microdeletion mutations that clinicians need to develop new treatments. Classifiers, trained using limited set of known mutated samples, are not capable of exact identification of mutations and their distribution within the sample, especially for previously unknown mutations. Consequently, results obtained by classification, are not reliable to select therapy in personalized medicine. Personalized medicine demands exact therapy, which can be designated only if all combinations of EGFR gene exon mutations are known. We propose computing system/model based on two modules: The first module includes training of knowledge based radial basis (RADBAS) neural network using training set generated with combinatorial microdeletion mutations generator. The second module has two modes of operation: the first mode is offline simulation including testing of the RADBAS neural network with randomly generated microdeletion mutations on exons 18th, 19th, and 20th; and the second mode is intended for application in real time using sample patients’ data with microdeletion mutations extracted online from EGFR mutation database. Both modes include preprocessing of data (extraction, encoding, and masking), identification of distributed mutations (RBNN encoding, counting of exon mutations distribution and counting of EGFR gene mutation distribution), and standard reporting. The system has been implemented in MATLAB/SIMULINK environment.