<<
>>

Graphical model of recognition of installation


For forecast interpretation make thermal cards with use dass aсtivatiοn mappings (CAM) [75].

For generation CAM it is required to load a picture into completely trained network and to gain result definitive svertochnym a layer.

Pustotobrazhenie functions, - weight in last layer

Classifications for function representation k, the leader to a pathology with. Card Mcнаиболее of the characteristic signs used at classification of the image, having a pathology is gained with.

Then:

Thus, the most important functions used by model in its forecast of a pathology c, by scaling of representation Mcдо of sizes of the image and the subsequent superposition of the image are defined.

Inoculated Dexnet localises pathologies, which identifitsiruyojutsja by means of cards of activation of the class-room which, in turn, gate out X-radiation areas, the most important for classification konkretyonyh pathologies.

However for raise of accuracy of the developed algorithm, capable to diagnose pulmonary diseases, there was an idea of its adaptation of localisation potalogy under more modern X-ray images of bodies - pictures kompyojuternoj tomographies.

Β algorithm ChexNet following changes have been made. After the analysis and research of new function of activation ReLU (abbreviation from English ^otified linear unit) which allows to speed up essentially process of instruction and simultaneously with

It considerably to simplify scalings (for the account of simplicity of the function), that means the block of linear rectification computing function f () = max (0,).

As of 2017 this function and its modifications (Noisy ReLU, Leaky ReLU and others) are most often used function of activation in deep nejrosetjah, in particular svyortochnyh.

Thus, the inoculated algorithm has been tested in pictures of a computer tomograph of pulmonary shares of the person. Instruction was made on the processed images by the permission 224? 224 pixels. The sample volume on which basis instruction was made, has made 15000 images from 50 patients. For instruction of a neural network the method of return extending of an error which consisted in kernel turn on 180 grades and scanning process svertochnoj cards of deltas was used so that the window of scanning fell outside the limits the image. Β research process it has been installed, that increases in quantity of nerve cells in the latent layer considerably are increased by number of iterations which are required for instruction of an artificial neural network. Minor improvement of results of work of algorithm is thus gained.

Β result of work of a neural network the processed image of a computer tomograph on which the thermal card with zele - nymi / the yellow/red tinctures specifying in the positive forecast of disease and dark blue/violet is put is gained - if disease on an observed section of picture KT is not revealed. Thus, the algorithm allows osushchestvyoljat the operative analysis given KT lungs and assumes definition kolicheyostvennoj estimations of probability of possibility of pulmonary disease. Instances reyozultatov algorithm works are presented in drawing 3.12

Drawing 3.12 - the Initial picture ΚΤ a lung and result of its machining svertochnoj a neural network

For check and testing of the offered algorithm the baseline of images of one of medical institutions of Voronezh has been used.

The applied baseline includes images ΚΤ with various aspects of formations: a pneumonia, tuberculosis, tuberkulemy, abscesses, emphysemas, sarkoidozy, a cancer (central and peripheral), metastasises.

On sample of 50 images ΚΤ the algorithm in 95 % of cases unmistakably defines disease and detects formations in a picture.

On the remained 5 % in the core cases when parts the bronchial tube or the big vessels wrongly are accepted for formations have.

The formations which are referring to to a diaphragm, thorax or sredosteniju it is difficult enough konturizovat.

The given problem originates in an aspect of that classification is made for each of contours individually. Therefore consideration not only separate contours, but also their sets considerably will raise efficiency of the offered method. Further the analysis becomes considerable martempering
Interframe changes of pulmonary drawing. Now the given analysis is made by doctors-radiologists at the description of the investigated image of a computer tomograph.

<< | >>
Scientific source Vasilchenko Vladislav Alekseevich. INTELLECTUALIZATION of PROCESSES of ADOPTION of MEDICAL SOLUTIONS Within the limits of a bioengineering system of DIAGNOSTIC And TREATMENT of PULMONOLOGICHESKY DISEASES. The dissertation on competition of a scientific degree of a Cand.Tech.Sci. Voronezh - 2019. 2019

Other medical related information Graphical model of recognition of installation:

  1. 5.3. Modelling of intergene interactions at patients bronyohialnoj an asthma the Buryat and the Russian living in Transbaikalian edge
  2. survey of mathematical models of an eyeball
  3. working out of mathematical electrophysical model and demands to boundary coditions
  4. single-layer flat mathematical electrophysical model of an eye
  5. two-layer flat mathematical electrophysical model of an eye for an estimation of agency of characteristics of a century on results of researches
  6. working out of the constructive solution of electrode system for conducting reooftalmograficheskih researches
  7. survey of mathematical methods of forecasting, feature of use of illegible logic of decision-making at urolithic sickness
  8. Methods of comparison of peak and phase data of an electrocardiogram
  9. the Maintenance
  10. Working out of a way of representation of the digitized video shot as uniform histogram of the image
  11. 3.6 Working out of system model of recognition of condition PZH and algorithm klasterizatsii the foreground of video shot TAUZI and TRUZI