Чрескожная криоабляция является перспективным методом лечения больных с метастатическим поражением костей. Грудина — одна из сложных локализаций для проведения процедуры в связи с высоким риском повреждения структур средостения и легких.ЦЕЛЬ РАБОТЫРассмотрение методологических аспектов проведения миниинвазивной криоабляции под контролем КТ при метастатическом поражении грудины, уточнение и систематизация пункционных чрескожных стереотаксических доступов. Процедура криоабляции была выполнена 9 пациентам. Размеры метастазов варьировали от 7 до 56 мм в наибольшем измерении. Несмотря на индивидуальную анатомическую изменчивость, при всех процедурах удалось достичь цели вмешательства и избежать осложнений. По результатам проведенных процедур были предложены оптимальные пункционные доступы в зависимости от объема опухолевого поражения грудины — фронтальный и латеральный. После завершения криоабляции опухолей грудины рекомендуется контрольная КТ грудной клетки для исключения внутриплеврального кровотечения.

This paper presents a review of methods for explaining and interpreting the classification results provided by various machine-learning models. A general classification of the interpretation and explanation methods is given depending on the type of interpreted model. The main approaches and examples of explanation methods in medicine and, in particular, in oncology, are considered. A general scheme of the explainable intelligence subsystem is proposed, which allows explanations using natural language.

Percutaneous image-guided cryoablation (PICA) in relieve pain from metastatic bone disease was performed in 24 patients. The cryoprobes were navigated and the ice sphere was monitored using computed tomography. Each patient, depending on the volume of the lesion, underwent from one to four ablations. Pain assessment was performed using a numerical rank scale. A decrease in the pain index after PICA was from 4.1±0.8 points to 1.3±0.5 points. The duration of analgesic affect correlated with local control of the tumor. A later postoperative complication was noted in one case in the form of a bone fracture in the ablation zone. Further study of the effectiveness of this technology is necessary by analyzing larger samples and comparing them with the results of treatment with other methods of local exposure.

Two new survival models, the deep survival forest and the Elastic-Net-Cox Cascade, are presented in the paper. They can be regarded as a combination of random survival forests and the Elastic-Net-Cox models with the deep forest (DF) proposed by Zhou and Feng. The main ideas to construct the models are to replace the original random forests incorporated into the DF with the corresponding survival analysis models. A stacking algorithm implemented in the deep survival forest and the Elastic-Net-Cox Cascade, which can be regarded as a link between the DF levels, uses quantiles of the random time-to-event and the mean time-to-event computed from the estimated survival functions at every level of the DF. Numerical examples with real data illustrate the proposed models.

This chapter reviews the basics and recent researches of computer-aided diagnosis (CAD) systems for assisting neuroradiologists in the detection, monitoring and prediction of multiple sclerosis (MS) in magnetic resonance (MR) images. The CAD systems consist of image feature extraction based on image processing techniques and machine learning classifiers such as linear discriminant analysis, artificial neural networks, and support vector machines. We introduce useful examples of the CAD systems in the neuroradiology and conclude with possibilities in the future of the CAD systems for MS in MR images.

This paper presents a prototype of an intelligent system for advanced analytics for integrated security of complex information and cyberphysical systems with the implementation of analytical models and software developed in Peter the Great St. Petersburg Polytechnic University . The article discusses the practical aspects of the application of unsupervised machine learning methods to the tasks of identifying abnormal objects in the field of information security in computer networks. The format of presenting initial data on various events in computer networks is described, as well as the process of preparing a training set for machine learning. The results of detecting anomalies by the Isolation Forest and Local Outlier Factor methods are presented, as well as an analysis of the results.

A new adaptive weighted deep forest algorithm which can be viewed as a modification of the confidence screening mechanism is proposed. The main idea underlying the algorithm is based on adaptive weigting of every training instance at each cascade level of the deep forest. The confidence screening mechanism for the deep forest proposed by Pang et al., strictly removes instances from training and testing processes to simplify the whole algorithm in accordance with the obtained random forest class probability distributions. This strict removal may lead to a very small number of training instances at the next levels of the deep forest cascade. The presented modification is more flexible and assigns weights to instances in order to differentiate their use in building decision trees at every level of the deep forest cascade. It overcomes the main disadvantage of the confidence screening mechanism. The proposed modification is similar to the AdaBoost algorithm to some extent. Numerical experiments illustrate the out-performance of the proposed modification in comparison with the original deep forest. It is also illustrated how the proposed algorithm can be extended for solving the transfer learning and distance metric learning problems.

e15519 Background: Metronomic therapy (MT) is one of the most promising advances in cancer treatment strategy focused on both tumor cells and their microenvironment, specifically anti-angiogenic, immune and metabolic regulation. The treatment modality is a durable cytostatic drugs exposure with less toxicity. We investigated the efficacy of MT in point of prolonged duration of response (more than 6 months) in pts with the advanced resistant tumors. Methods: A total of 678 pts have been enrolled in the study and have been treated MT cyclophosphamide 50 mg (qd, PO) and methotrexate 2.5 mg (BID, PO) twice per week dosing schedule until progression. 508 pts (75%) had disease progression during first 6 months. Long-term efficacy (more than 6 months) was assessed in 170 pts (25%). The study population characteristics were female 122 (72%) vs male 48 (28%). Among the enrolled pts median age was 70 years (range from 37 to 91 years). ECOG status ranged from 0 to 4 (median - 1). The majority of pts received MT as a 2nd and 3rd line (57%) of the therapy (range from 1 to 11): 1-2 – 106 (62%), 3-4 - 52 (31%), 5 and more - 12 (7%). The distribution of pts by diagnosis is presented in the table 1. The majority of MT pts were represented by advanced clinical stage III-IV (97,6%): I - 1 (0,5%), II - 3 (2%), III - 7 (4%), IV - 159 (93,5%). MT commonly used in early-stages cancer in elderly patients with multiple comorbidities contraindicated to standard treatment. Most pts had two or multiple metastatic sites. Results: The ongoing long-term effect for MT indication was from 6 to 46 months. The main group consisted of the breast and colorectal cancer, 27% and 15%, respectively. The median time to progression was 9 months (16 months in CR was achieved (0,4%), 11 months and 9 months with a partial response (3,4%) and stable disease (21,2%), respectively). 12 (1,7%) pts are on ongoing long-term follow up more than 24 months. MT was well-tolerated. Drug related adverse events were described upon the study: hematological (10,2%, Gr 3-4 – 3,5%) vs non-hematological (6,6%, Gr 3-4 1,7%) toxicity. Conclusions: MT schedule demonstrated effectiveness in patients with advanced resistant tumors. The median duration of the effectiveness doesn’t depend on the line of therapy, tumor localization and objective response. The role of possible predictive markers of clinical benefit is question of further trials. [Table: see text]

A very simple algorithm for explaining decisions of cancer computer-aided diagnosis systems is proposed. The algorithm produces explanations of diseases in the form of special sentences via natural language. It consists of two parts. The first part is implemented by using a standard local post-hoc explanation model, for example, the well-known method LIME. This part aims to select important features from a segmented and detected suspicious object or its low-dimensional feature representation. The second part is a set of simple classifiers which transform the selected important features into one of the classes corresponding to simple phrases. The explanation sentences in natural language are composed of the simple phrases (primitives) such that every subset of the phrases describes a single peculiarity of a suspicious object, for example, a shape, a structure, inclusions, contours. The primitives are regarded as classes for the set of classifiers. The proposed algorithm is general and can be applied to implementing the explanation subsystem of various diseases.

A new meta-algorithm for estimating the conditional average treatment effects is proposed in the paper. The basic idea behind the algorithm is to consider a new dataset consisting of feature vectors produced by means of concatenation of examples from control and treatment groups, which are close to each other. Outcomes of new data are defined as the difference between outcomes of the corresponding examples comprising new feature vectors. The second idea is based on the assumption that the number of controls is rather large and the control outcome function is precisely determined. This assumption allows us to augment treatments by generating feature vectors which are closed to available treatments. The outcome regression function constructed on the augmented set of concatenated feature vectors can be viewed as an estimator of the conditional average treatment effects. A simple modification of the Co-learner based on the random subspace method or the feature bagging is also proposed. Various numerical simulation experiments illustrate the proposed algorithm and show its outperformance in comparison with the well-known T-learner and X-learner for several types of the control and treatment outcome functions.

The main difference between artificial intelligence (AI) systems and simple automated algorithms is the ability to learn, synthesize and conclude. The AI system is trained on a set of examples, including pictures, characteristics of patients with a certain disease, then it allows to generalize a lot of such examples and get some general functional dependence, which brings in line the patient data and a certain diagnosis. The system can be named intelligent if this synthetizing ability is realized. Although the AI systems are now becoming more understood and accepted by doctors, a deeper understanding of «how it works» is needed. The article provides a detailed review of the application of methods and models of artificial intelligence in the diagnostics of cancer based on the of multimodal instrumental data. The basic concepts of artificial intelligence and directions of its development are presented. From the point of view of data processing, the stages of development of AI systems are identical. The stages of intellectual processing of diagnostic data are considered in the paper. They include the acquisition and use of training databases of oncological diseases, pre-processing of images, segmentation to highlight the studied objects of diagnosis and classification of these objects to determine whether they are malignant or benign. One of the problems limiting the acceptance of AI systems development by the medical community is the imperfection of the explainability of the results obtained by intelligent systems. Authors pay attention to importance of the development of so-called explanatory intelligence, because its absence currently significantly inhibits the introduction and use of intelligent diagnostic systems in medicine. In addition, the purpose of the article is a way to develop the interaction between a radiologists and data scientists.

Aim: study of the predictive value of determining ctDNA during treatment with osimertinib in patients with NSCLC with EGFR mutation. Methods: The study included patients with metastatic EG-FR-associated NSCLC, in whom, with progression against the background of 1st - 2nd generation TKIs, the T790M mutation was detected. Patients received osimertinib therapy 80 mg/ day, daily, until progression. Before treatment, and then every 2 months, whole blood was taken to conduct a qualitative assessment of ctDNA in dynamics by the RT-PCR method. Results: From 2016 to 2019 in St. Petersburg Clinical Scientific and Practical Center of Specialized Types of Medical Care (Oncology), 22 patients were identified T790M associated progression of EGFR NSCLC. 81.9% (18/22) are women, 18.1% (4/22) are men. The average age is 61.2 years (50-75). 1/22 had smoking experience for more than 30 years. The molecular genetic profile in 16 is represented by ex19del, 5 L858R, 1 -a combination of rare mutations G719S+S768I. The effect of therapy was evaluated in 20/22 patients. PR and SD were registered in 9/20 (45%) and 10/20 (50%) patients, respectively. Median PFS - 16.7 months (cI 95%, 11,4-22,0). In 12/22 patients was observed the disappearance of ctDNA T790M after 2 months of osimertinib therapy. PFS is 18,9 months (95% CI, 14,8-19,7), in patients with no mutation detected in the second month of treatment compared with the group of patients in which the ctDNA was determined (PFS 8.0 months) (CI 95%, 4,2-11,8) (p=0.015). Correlation analysis did not reveal any clinical factors associated with the disappearance of ctDNA. Conclusions: The disappearance of ctDNA in plasma after 2 months of treatment with osimertinib is associated with an increase in PFS and can be considered as a predictive marker in patients with metastatic NSCLC EGFR T790M.

Two algorithms for explaining decisions of a lung cancer computer-aided diagnosis system are proposed. Their main peculiarity is that they produce explanations of diseases in the form of special sentences via natural language. The algorithms consist of two parts. The first part is a standard local post-hoc explanation model, for example, the well-known LIME, which is used for selecting important features from a special feature representation of the segmented lung suspicious objects. This part is identical for both algorithms. The second part is a model which aims to connect selected important features and to transform them to explanation sentences in natural language. This part is implemented differently for both algorithms. The training phase of the first algorithm uses a special vocabulary of simple phrases which produce sentences and their embeddings. The second algorithm significantly simplifies some parts of the first algorithm and reduces the explanation problem to a set of simple classifiers. The basic idea behind the improvement is to represent every simple phrase from vocabulary as a class of the "sparse" histograms. An implementation of the second algorithm is shown in detail.

Artificial intelligence (AI) in the fourth industrial revolution is integrated into the life of modern society. Modern computer technology and software are opening the way for implementation of AI also in medicine. In fact, the position of AI in relation to medicine was unacceptable by the medical community a few years ago. Nevertheless, it is obvious today that the development of IT, including AI, has an impact on the quality of medical care, particularly of diagnosis. Integration of AI systems as an organisational innovation in the work of medical institutions is a significant challenge. It involves the revision of some aspects of the professional education and of issues of the interdisciplinary interaction. A scheme of effective cooperation of specialists of IT and radiologists is proposed in the paper. It aims at the permanent development of AI systems and their implementation in a daily medical practice and proposed by an example of the medical intellectual computer-aided diagnostic system Doctor AIzimov.

The relevance of developing an intelligent automated diagnostic system (IADS) for lung cancer (LC) detection stems from the social significance of this disease and its leading position among all cancer diseases. Theoretically, the use of IADS is possible at a stage of screening as well as at a stage of adjusted diagnosis of LC. The recent approaches to training the IADS do not take into account the clinical and radiological classification as well as peculiarities of the LC clinical forms, which are used by the medical community. This defines difficulties and obstacles of using the available IADS. The authors are of the opinion that the closeness of a developed IADS to the «doctor’s logic» contributes to a better reproducibility and interpretability of the IADS usage results. Most IADS described in the literature have been developed on the basis of neural networks, which have several disadvantages that affect reproducibility when using the system. This paper proposes a composite algorithm using machine learning methods such as Deep Forest and Siamese neural network, which can be regarded as a more efficient approach for dealing with a small amount of training data and optimal from the reproducibility point of view. The open datasets used for training IADS include annotated objects which in some cases are not confirmed morphologically. The paper provides a description of the LIRA dataset developed by using the diagnostic results of St. Petersburg Clinical Research Center of Specialized Types of Medical Care (Oncology), which includes only computed tomograms of patients with the verified diagnosis. The paper considers stages of the machine learning process on the basis of the shape features, of the internal structure features as well as a new developed system of differential diagnosis of LC based on the Siamese neural networks. A new approach to the feature dimension reduction is also presented in the paper, which aims more efficient and faster learning of the system.

A modification of the confidence screening mechanism based on adaptive weighing of every training instance at each cascade level of the Deep Forest is proposed. The modification aims to increase the classification accuracy. It is carried out by assigning weights to training instances at each forest cascade level in accordance with their classification accuracy. Larger values of accuracy produce smaller weights. Two strategies for using the weights are considered. The first one when the weights are regarded as probabilities of choosing the corresponding instances in building decision trees. According to the second strategy, the weights are used in splitting rules. The modification increases the classification accuracy and may reduce the training time for many real datasets. Numerical experiments illustrate good performance of the proposed modification in comparison with the original Deep Forest proposed by Zhou and Feng.

Robust weighted aggregation schemes taking into account imprecision of the decision tree estimates in random forests and in random survival forests are proposed in the paper. The first scheme dealing with the random forest improves the classification problem solution. The second scheme dealing with the random survival forest improves the survival analysis task solution. The main idea underlying the proposed modifications is to introduce the tree weights which take simultaneously into account imprecision of estimations as well as aims of the classification and regression problems. The imprecision of the tree estimates is defined by means of imprecise statistical inference models and interval models. Special modifications of loss functions for the classification and regression tasks are proposed in order to simplify minimax and maximin optimization problems for computing optimal weights. Numerical examples illustrate the proposed robust models.