RO133032A2 - Support infrastructure for intelligent imaging diagnostic - Google Patents

Abstract

The invention relates to a computer-assisted diagnostic system which is able to help doctors for a more rapid and reliable interpretation of the recordings delivered by an endoscopic videocapsule, combining said information with that obtained from a computerized tomographic examination, being intended, in particular, for the digestive tract pathology. According to the invention, the system consists of a mobile data input device (4), held by a doctor, and a central information system (6) having a data base (8), a set of modules for imagistic processing and an assembly of complex neural networks (9), where the system (6) takes over a patient's data and the files resulting from the imaging investigations remotely delivered by the doctor by means of a graphical interface (5) of the mobile device (4), stores them in the data base (8), wherefrom the set of modules for imagistic processing and the neural networks (9) take them over and process them, giving the doctor a set of imaging sequences containing digestive tract lesions together with their classification.

Description

DESCRIPTION OF THE INVENTION; i.> tie tiepoziț .....

"SUPPORT INFRASTRUCTURE FOR DIAGNOSIS

INTELLIGENT IMAGISTIC ”

The invention generally relates to an advanced computer-assisted diagnostic (CAD) system capable of assisting physicians in the process of faster and more reliable interpretation of endoscopic video recorder (VCE) records, combining this information with data obtained from computed tomography (CT) examination.

The system will be based on different image recognition techniques used to isolate lesions and remove potential artifacts with similar characteristics from the analysis, taking full advantage of the collaborative aspects of telemedicine applications, web interfaces and portable devices. The output data of the algorithms will represent inputs to a set of multi-layered artificial neural networks (ANNs) capable of self-improvement. The entire set will be designed as a cloud-based, stationary system, but fully integrated into the World Wide Web and other communication networks, designed so that a wide variety of data will be used in the early stages to train the system to complete with success of his tasks. The second component of the system will combine the imaging data extracted from the computed tomography records and the imaging data acquired by VCE in real time from the small intestine, thus obtaining a three-dimensional exploration through the small intestine or even through the upper / lower segments of the digestive tract. , providing additional help for clinicians.

The system consists of a mobile device for recording and transmitting data to a central unit, which performs the classification of the lesions present at the level of the digestive tract, following processing with the help of specific techniques of image processing, as well as of learning based on neural networks. , combining information obtained from VCE and CT investigations. All the parameters are taken in real time by the clinician, the system itself being considered as "cloud-computing" type; data entry is done remotely, through a graphical interface, using the dedicated mobile device, permanently connected to Intemet.

The system can be used to identify the pathology of the digestive tract in the context of establishing a diagnosis in a classic consultation, but also within telemedicine applications, being an expert reference system for remote centers. Also, it can be used both in the training activities of medical professionals and in specialized clinics, because the relative ease of use and the interactive approach make it a useful tool in the teaching process for the medical personnel in the higher education.

Through the mobile unit interface, the medical framework introduces the clinical information corresponding to the patient, to which is added the data from the imaging investigations (VCE and CT) taken in electronic format following the appropriate investigations. Subsequently, the module sends the data through a permanent Intemet connection (mobile data subscription or existing wireless network) to a central computer system, where they are entered into a database, from where the whole neural network processes the information and provides the doctor with an answer specific.

to 2017 00441

03/07/2017

The invention proposes the integration of an advanced imaging analysis system assisted by a ANN-based CAD module, into a complete integrated web platform, available in a variety of forms, designed to address a specific challenge, represented by the correct classification and interpretation of the recordings. VCE. Multiple attempts have been made to integrate software solutions and to help physicians identify the lesions present in the images obtained by VCE and CT, which sum up hours of viewing; however, as far as we know, this is the most advanced and coherent approach to this issue. The system described in this invention will be available for a variety of mobile platforms - tablets, smartphones, as well as mobile or stationary workstations (laptops or dedicated computers), all of which are connected to the mainframe that will have the interface also unified by the back-end processes required for imaging analysis and diagnostics assisted by ANN. This product will be a very useful tool for physicians throughout the medical system - from large hospitals, to smaller clinics, which do not have the expertise to interpret VCE, providing independent help in classifying meaningful images. A second use of this system will be for training purposes for young doctors, as an integral part of their medical education, as it will assist in making decisions and provide useful and quick indications for a successful diagnosis. In addition, the second component represented by virtual enteroscopy, which combines the volumetric data obtained at CT with the imaging aspect of the texture of the intestinal mucosa provided by VCE, is an interesting new tool that can be used for teaching and training purposes, both in university education. , as well as in the post-university one.

The small intestine is currently the most difficult to examine segment of the digestive tract, due to its location distal to both natural orifices, commonly used during routine digestive endoscopy. However, examination is vital because a significant number of pathologies, most of them malignant with vital patient implications, are located at this level. Standard diagnostic techniques are limited to barium radiological investigations, push-enteroscopy and intraoperative enteroscopy, selective arteriography, scintigraphy or computed tomography. All of these methods are either too invasive or do not provide enough data for accurate diagnosis.

The endoscopic videocapsule, together with balloon enteroscopy (single or double), or spiral are considered two of the most valuable means of investigation available to the gastroenterologist. VCE is currently the optimal method of exploration of the small intestine, due to its ease of use and relatively low cost.

The VCE system consists of a cylindrical device measuring approximately 26x11 mm and weighing approximately 3 grams. The on-board image sensor captures 4 to 14 images per second. The device also contains an array of LEDs, optical circuits, an ASIC control chip, an RFID antenna system, and a battery that runs for approximately 8 hours (average duration of a full GI transit). The resulting film contains over 50,000 images, which are transmitted to an external recording device equipped with a series of electrodes, which is easy to carry, allowing the patient to perform a series of simple activities. The recordings are taken for further analysis with the help of high performance workstations, with the possibility of being viewed on a small LCD screen attached to the device.

Subsequently, an experienced clinician needs two to four hours to analyze the video recordings, depending on his experience and the nature of the suspected pathology. By default, it is necessary to develop systems capable of performing the automatic analysis of 2017 00441

03/07/2017 images and provide support in the possible detection and classification of lesions. Thus, the medical framework needs an intelligent system, capable of analyzing the whole VCE image, to identify a series of modified frames and to classify potential lesions, while assisting the investigator in exploring the digestive tract. The combination of these images with those obtained from the CT investigation provides more precision in establishing the diagnosis, and a new perspective on the digestive tract. A number of image processing methods and artificial intelligence techniques will be used for the purpose of automatic segmentation of frame content and for classification of lesions.

The neural network represents a set of decision structures called neurons, the connections of which are modulated during a series of consecutive stages, namely training, validation and testing. The layers formed by these neurons fall into three categories: input, decision and output. In the input layer, the data in numerical format are entered in the system, where each parameter is given a weight, which ultimately counts for the final result, following the nonlinear calculation based on the respective weights. During the training process, the weights are adjusted accordingly so that the result approximates the expected one.

The neural networks of applicative nature in medicine have been used in medical applications and in the past, in the diagnostic process, in the risk forecasting or in the evaluation of the prognosis of some diseases. The results were variable, depending on the type of parameters introduced, their quality and relevance, as well as the efficiency of the procedures used during the training. The neural networks used so far in medical studies and especially in the identification and classification of digestive tract pathology have presented a number of limitations:

the low number of images involved in the training process, which led to the selection of too specific variables, the cases being classified "ideally" in the model of diagnosis sought;

incomplete identification of the imaging parameters specific to the pathology followed; erroneous selection of the input data set, either too numerous, insufficient or not of clinical importance;

wrong estimation of error rate, resulting from improper validation; stopping the learning process after a small number of cases, thus depriving the system of the plasticity and adaptability that should have been the defining ones.

The technical problem addressed by the present invention is the improvement of the diagnostic method assisted by an expert system based on a set of neural networks, applied to the pathology of the digestive tract, especially the small intestine, capable of combining different techniques of imaging exploration in order to increase the accuracy of lesion detection.

This purpose is achieved by the following specific characteristics:

Input data can be entered in real time, from any location where Internet access exists, allowing virtually any use of the system in almost any existing clinical context;

the data can be entered in real time, by using the dedicated device;

the validation can be done at any time by human operators, for the purpose of 2017 00441

03/07/2017 (2y eliminating the possibilities of incorrect estimation of the error rate;

the parameters will be stored in a dynamic database, from which they can be used subsequently for successive training, after validation of the diagnoses by an experienced clinician; In this way one of the main drawbacks of classical neural networks is eliminated, namely the stopping of the active learning process at a certain level; the number of cases present in the system is practically unlimited, with any reference medical centers being able to offer new results for system improvement at any time; data recording and processing will be performed centrally, within the system dedicated for this purpose, located at a distance;

using imaging investigations performed, by converting them into numerical values specific to the visual characteristics of the lesions possibly encountered in the digestive tract;

the system allows both the individual analysis of the digestive tract images, offering a point response, but also the analysis of the entire film resulting from the VCE investigation, for a complete answer;

the modularity of the system allows it to provide a result even in the absence of the CT investigation, by analyzing the available images and determining an established result with a certain degree of error; the system is thus flexible and eliminates the subjectivism of a human operator.

The following is an example of embodiment of the invention in relation to Figures 1 and 2. Figure 1 is a block diagram of the automated mobile system based on neural networks for the identification and classification of lesions present in the digestive tract.

The operator (landmark 0) acquires, through the VCE system (1) and CT (2), the specific images of the digestive tract, which are transferred using a mobile transfer device (3) to the mobile device (4). Through the graphical interface (5), the operator (0) can select either the entire film or sections thereof for further processing. The mobile system 4 sends the data through a secure wireless connection to the central unit 6, a personal computer connected to the Internal, with a unique address, which contains the data reception application (7), the database (8) and the neural network assembly. (9). The database contains both the original images and the numerical values that encode selected image parameters.

to 2017 00441

03/07/2017

Fig. 1

Figure 2 shows a diagram of the workflow that emphasizes the successive steps required for a complete image analysis. The distance between the original VCE film along with the CT investigations, and the set of lesions identified and classified from those sequences cannot be covered in one step. The processing sequences are complex and use several steps that lead the doctor closer to the expected result. The general algorithm consists of several major stages: analyzing each VCE framework, classifying it as a usable image (whether it provides relevant information or not), eliminating potential misleading artifacts (bubbles, food scraps), determining and extracting the specific features of each frame, using an intelligent system for the classification of individual cadres; if a particular lesion is identified, it is necessary to analyze the following frameworks to determine if the same lesion appears in several stages; subsequently, an overall analysis is performed to reconstruct the lesion and classify it. The result of the algorithm is a collection of reconstructions of the potential lesions identified and classified previously, in order to help the examining doctor in establishing a diagnosis.

The first stage receives as input data the original VCE film and CT images and provides a by-product represented by a set of usable frames. All subsequent stages are linked, the result of one stage constituting input data for another stage. The final stage provides a set of potential lesions, consisting of the graphic representation of the lesions, the identifiers of the images that contain them, their classification and the probability of classification. Each important step can constitute a dedicated work package, these are designed to receive input data and provide independent output data.

The proposed architecture targets a system with increased connectivity and interoperability using HL7 standards. The main objective is to design a computer system capable of integrating and combining different stages, implemented as individual modules, in a centralized application that facilitates the accomplishment of a single task and promotes the proper reuse and integration of information.

to 2017 00441

03/07/2017

A description of how the method works can be found in the following:

the doctor examines the patient by conventional methods; thereafter, the examination is performed with the endoscopic video capsule, as well as the tomographic computer investigation, according to the classical protocols;

upon completion of the investigations, the recordings are retrieved on the mobile device using a mobile recording device (for example a USB stick);

the doctor uses the mobile device to enter the data, then processes the obtained film, with the potential determination of the sequences of interest, if it is not desired to use the entire film. The mobile device extracts the relevant data and processes it for transmission to the central system;

the data is transmitted remotely to the central system, where they are processed;

the whole neural network takes over the significant data and in a few seconds a result is offered;

the doctor receives a potential identification and classification of the lesions encountered in the digestive tract, with a certain degree of safety;

the doctor establishes a personal diagnosis, which will later be used in a new procedure for training the whole neural network, if the diagnosis is confirmed.

Claims (1)

Semi-automatic system based on a set of neural networks for the identification and classification of the digestive tract pathology, especially the small intestine, characterized by the fact that it includes a mobile unit (4) for entering the data corresponding to the patient, together with the numerical data resulting from the analysis. images provided to the doctor after performing the imaging investigations, a central computer system (6) containing an information retrieval interface (7), a database (8) and a set of neural networks (9). The doctor offers an immediate response through the same mobile device, the answer being also saved in the database (8), the doctor introducing his own diagnosis that re-trains the system offering new weights for the new data sets introduced globally in the system.