Release date: 2016-09-22
From the beginning of the question to the medical image
Under normal circumstances, organs and tissues in the human body cannot be seen with the naked eye. In ancient times, famous doctors such as Bianque and Huaying diagnosed the internal causes of patients by "looking, smelling, asking, and cutting". This is the most "advanced" diagnosis of that era.
One day in 1816, French doctor Reneck walked down the street, accidentally seeing a few children hitting one end of a piece of wood with a big nail, while the other children used their ears to stick to the other end of the wood. To listen to the sound, this gave Dr. Reneck a great inspiration. When he got home, he immediately found someone to make a hollow wood tube, which was the first stethoscope in human history. Later, the stethoscope was widely used in the heart and obstetrics.
French doctor Renek auscultated with a wooden tube
In modern times, doctors observe the condition of the patient's body and no longer just use a stethoscope to solve it. In 1971, the advent of CT marked the official establishment of medical imaging. With the advancement of medical imaging technology, the medical imaging department evolved from radiology became the fastest-growing discipline in clinical medicine, expanding from traditional X-ray examination. To ultrasound, radionuclide imaging, X-CT, MRI, digital imaging, and today's most advanced PET-CT technology. With these new technologies, doctors are able to "snoop" the lesions inside the body more deeply.
Fusion Master of Impact Data - PACS System
The emergence of medical imaging equipment has made medical clinics increasingly rely on medical imaging inspections. Traditional medical image management methods (films, pictures, and materials) are accumulated over the years, stored and kept year after year, and pile up mountains, which brings many difficulties to find and read. The loss of movies and materials occurs in hospitals. The traditional method of document management has been unable to meet the management requirements for such large and large-scale medical images in modern hospitals.
With the development of database technology and computer communication technology, digital image transmission and electronic film came into being. Many hospitals have carried out hospital information reforms. With the gradual update of imaging equipment to digitalization and the maturity of the Internet, filmless radiology and digital hospitals have become a reality. Regarding electronic film, we will introduce it in detail in the next article. I will not elaborate on it for the time being.
In order to uniformly store and manage the informationized data of different medical imaging equipments, the PACS system, the master of data integration of each platform, was born.
The English translation of the PACS system is the meaning of the image archiving and communication system. Its main task is to pass various medical images (including nuclear magnetic, CT, ultrasonic, various X-ray machines, various infrared instruments, microscopes, etc.) to various interfaces (analog, DICOM, The network) is stored in a digital way. When doctors need them, they are quickly transferred back to use as a housekeeper, perfectly acting as a lubricant between instruments.
PACS system application diagram
For a complete PACS system, the main functions consist of three aspects: first, image acquisition, second, data transmission and storage, and third, image analysis and processing.
There are three main ways to collect images: pure digital acquisition, video capture, and film scanning.
In terms of information storage, the PACS system uses two different methods for storing structured data and unstructured data. Use a database to manage structured data such as patient information, and use a file system to manage unstructured data such as image data. It is like a person carrying a baggage to take a plane, the baggage is checked into the baggage compartment, and the person is sitting in the cabin, and the two do not interfere with each other.
In addition, because the data files of medical images tend to be large, the conventional CT scan is of the order of 10MB, the chest radiograph of the X-ray machine can reach 20MB, and the image of the cardiovascular image can reach more than 80MB. The traditional method is generally to use the server and the optical disc for storage, which is relatively rigid and difficult to expand. At present, the emerging cloud computing cloud storage technology has the functions of fast data calling, network sharing and application development, and combined with the PACS system, it will be a major direction of image storage in the future.
The principle is also very simple: the hospital deploys the PACS system to a third-party cloud platform, and realizes all-weather image storage through the distributed and load-balanced cluster system of the cloud platform. The establishment of the cloud platform can also achieve a full integration of cross-platform, multi-terminal, PC and mobile devices, thereby completely achieving imageless paper, no disc, and no film.
This new model not only enhances the efficiency of each doctor, the quality of work, but also enriches the collaborative work scene of doctors. In addition, the hospital does not have to spend a lot of money to purchase the server, thereby reducing the cumbersome post-maintenance and expansion to achieve cost-saving purposes.
OK! The problem of data storage has been solved, but the standardization of data has become a new problem. Although hospitals can use the PACS system to achieve information interoperability between various types of instruments, the data collection and transmission are very difficult because the data standards used by different manufacturers' equipment and different PACS systems are different. It’s like people from different languages ​​and countries have met together. You talk about your ABC. I said that I have eaten. How to make the products of these different countries and different manufacturers form a unified standard has become the biggest obstacle.
In this regard, Americans are always at the forefront of the times. In 1985, the American Society of Radiology ACR and the National Electrical Manufacturers Association NEMA jointly developed a standard for digital imaging and related information formats and information exchange methods: digital imaging and communications in medicine. Abbreviated as DICOM. The emergence of DICOM redefines the medical image format for clinical data exchange.
Under the DICOM standard, imaging equipment provides uniform standard image data to the PACS system. In terms of external communication, the PACS system still uses DICOM, which forms the maximum uniformity. To put it simply, it is to have a single interface for each instrument, just like we use English as the universal language of the world.
In 1993, DICOM successfully developed to the third generation, which is the DICOM 3.0 standard. As more and more medical device manufacturers in the country announce support for the DICOM 3.0 standard, DICOM 3.0 has become a recognized standard in the medical imaging industry worldwide.
The PACS system was originally used primarily in the radiology department. As a core component of the hospital's HIS system, the PACS system generally follows the HL7 standard and the IHE specification when constructing a hospital information system network. With the continuous improvement of the HL7 standard and the IHE specification, PACS has expanded from the simple image storage and communication between several radiographic equipments to the interoperability of all hospital imaging equipment and even different hospital images. Methods such as Mini PACS (micro PACS), departmental PACS, hospital-wide PACS, regional PACS, etc.
Mini-PACS: refers to only a single type of imaging equipment, CT or MRI.
Departmental PACS: Multiple imaging equipment in the radiology department can realize the sharing of images and diagnostic reports.
Hospital-wide PACS: Integrates clinical attending physicians, radiologists, and specialists in various departments of the hospital, as well as various imaging, medical records, and diagnostic reports.
Regional PACS: PACS network of the regional and inter-regional wide area network.
Schematic diagram of departmental PACS system
In short, the emergence of the PACS system not only solves the problem of image acquisition, but also solves the problem of data transmission and storage. As for the analysis and processing of the images that have not been mentioned, we will explain in detail later. Before that, let us first understand Under the medical image big data.
The causes of the formation of medical image big data in China
As a new term, whether it is medical image big data or medical big data first, it has not been verified. But to explain the medical image big data, it is necessary to make two points: one is the definition of medical image big data, and the other is the formation of medical image big data.
The definition of big data is a collection of data that cannot be captured, managed, and processed by conventional software tools within a certain time frame. It requires a new processing model to have stronger decision-making power, insight and process optimization capabilities to adapt to massive amounts. High growth rates and diverse information assets.
IBM summarized the 5V features of big data: Volume, Velocity, Variety, Value, Veracity.
Medical image big data, if defined by big data, is a large-scale, high-speed, multi-structure, high-value and true-accurate image generated by medical imaging equipment such as DR, CT, MR, etc. and stored in the PACS system. Data collection. It is equivalent to hospital information system (HIS) big data, inspection information system (LIS) big data and electronic medical record (EMR).
The two structures of multi-structure and high value are well understood and are structured and unstructured data with medical analysis and guiding value generated by the growing variety of medical imaging equipment. Large-scale and high growth rates need to be explained from the big environment.
There are two main reasons for the formation of medical image big data in China: one is the market, and the other is the population.
In terms of market size, as of June 2015, the number of top three hospitals in China was 705; CHIMA's data from 2014 to 2015 showed that the level of PACS, multi-disciplinary or hospital-level PACE systems in China has reached 60-70% respectively. And 50-60%, basically covering the top three hospitals in the first-tier cities in China.
From the perspective of market growth, China's PACS market has an average annual growth rate of more than 25%. According to ACMR survey data, from 2012 to 2015, China's PACS market continued to expand at a rate of more than 20%.
In terms of population, the main reason for the impact of medical image big data is the population base and age distribution. According to the main data bulletin of the sixth national census of the National Bureau of Statistics, the total population of the country is about 1.37 billion. From the growth rate and proportion of the elderly population, as of the end of 2014, China’s elderly population over 60 years old has reached 212 million, accounting for 15.5% of the total population. It is predicted that by the middle of this century, the number of elderly people in China will reach a peak of more than 400 million, and there will be one elderly person per three people.
Therefore, the current popularity and population of PACS systems is a large-scale basis for medical imaging big data in China; and the rapid growth rate of PACS systems and the elderly population is the basis for the high growth rate of medical image big data. Together constitute the cause of the formation of medical image big data in China.
As the last of the big data 5V features, how should the authenticity of medical image big data be realized? This involves data processing technology.
Data processing and "fishy pork"
In simple terms, the data collected by PACS systems from different imaging devices is often of varying quality. The degree of error and credibility of data analysis and output results depend to a large extent on the quality of the collected data. The so-called "garbage in, garbage out", without the accuracy guarantee of data, big data analysis becomes A piece of empty talk.
At present, the medical image post-processing methods mainly include two types, one is the direct processing technology port, and after the imaging examination of the patient, the software is directly used to process the image on the imaging device, for example, angiography on CT and MRI devices. Wait. The shortcomings of this method are obvious, and the image cannot be changed. The doctor can only rely on his own experience for pathological processing, which leads to inaccuracy of the data results.
For example, when CT images encounter mutual tissue imaging overlap, ordinary software image processing tends to interpret these overlapping data as noise or other interfering signals, while medical experts need to maintain the upper boundary or target contour of the image. The geometry of the boundary keeps the spinning constant (in short, to maintain the integrity of the image), which brings unpredictable difficulties to the doctor's diagnosis.
In addition to the processing of the imaging device software, there is also a method of transmitting image data to the PACS system through the image device, and the image is post-processed by the PACS system. For example, the PACS system uses multi-dimensional image fusion (CT/MRI/PET-CT) technology to segment, register and cluster images to preserve the authenticity of image data.
Multidimensional image fusion (CT/MRI/PET-CT) schematic
Multi-dimensional image fusion This "black technology" mainly includes data preprocessing, image segmentation, feature extraction and matching judgment. It may sound a little more cumbersome. In simple terms, data preprocessing refers to the vast amount of raw data from a variety of sources in a medical image database, with a large amount of ambiguity, incompleteness, noise and redundancy. Information. Therefore, before data mining, this information must be cleaned and filtered to ensure consistency and certainty of the data, making it into a form suitable for mining.
We are well aware that the medical image database contains a large amount of image data. For the sake of explanation, we compare these image data to various ingredients and compare the final processed information to the dish of fish-flavored pork.
Data pre-processing, you can think of it as the process of cleaning the ingredients. To make the dish of fish-flavored pork, you must first wash the pork, carrot, green pepper and even the onion and ginger, and filter out the residue, leaving The essence can do the next step. This stage, including image denoising, enhancement, smoothing, sharpening, etc., is collectively referred to as data preprocessing.
After the “food†is cleaned, it enters the image segmentation and feature extraction process. In this section, we can assume the process of “cutting†of the “foodâ€. Taking the domestic medical imaging company Huihui Huiying as an example, using the multi-dimensional image fusion technology, through the organ morphology model, the image edge feature model, and the neural network clustering model, the computer automatically divides the bladder, prostate, rectum, etc. of the pelvic CT automatically. (Division accuracy <2mm.), which provides the necessary image processing tools for later intelligent matching and judgment.
In the last step, we will process the “food ingredients†processed by the first two processes, and stir-fry the onion and ginger into a plate of fish-flavored pork, which is the process of image matching and clustering. The core technology that this stage of the PACS system relies on is deep learning, which is what we call artificial intelligence. Next, let's understand how artificial intelligence is applied to the field of medical imaging.
Inaccurate, large gap, artificial intelligence background
In August this year, the State Council issued a notice on the “13th Five-Year National Science and Technology Innovation Planâ€, and artificial intelligence has become a major focus. The "Planning" clearly points out that it focuses on the development of big data-driven human-like intelligent technology methods; breaks through the human-centered human-mass fusion theory and key technologies, and develops related equipment, tools and platforms; and human intelligence based on big data analysis. The direction has made important breakthroughs, realizing human-like vision, human-like hearing, human-like language and human-like thinking, supporting the development of the intelligent industry.
Before we explore how to apply artificial intelligence to medical imaging, we must first understand the two problems faced by medical imaging without artificial intelligence.
According to the arterial network, more than 90% of the medical data comes from medical images, but most of these data are subject to manual analysis. The shortcomings of manual analysis are obvious. The first is inaccuracy. It can only be judged by experience, and it is easy to misjudge. According to a misdiagnosed data from the Chinese Medical Association, the number of misdiagnosed cases per year in clinical care in China is about 57 million, the total misdiagnosis rate is 27.8%, the misdiagnosis rate of organ ectopic is 60%, and the average misdiagnosis rate of malignant tumors is 40%. Nasopharyngeal carcinoma, leukemia, pancreatic cancer, etc., the average misdiagnosis rate of extrapulmonary tuberculosis such as liver tuberculosis and stomach tuberculosis is also above 40%.
The second is the large gap. According to the data of the Arterial Network Eggshell Research Institute, the annual growth rate of medical image data in China is about 30%, and the annual growth rate of radiologists is about 4.1%. The gap between them is 23.9%. The growth in the number of radiologists is far less than the growth in image data. This means that radiologists will become more and more stressed in the future to process image data, even far exceeding the load.
Radiologist's work diagram
This point can be seen from the current work of radiologists. In January of this year, the medical profession had conducted a survey of 1,241 medical imaging doctors. One of the data is very noteworthy: more than 71% of the imaging doctors are looking forward to it. The return of radiation is false.
According to the data in the report, more than 50% of doctors work more than 8 hours, and 20.6% of doctors work more than 10 hours a day. Many doctors report that radiation is not known. Many doctors leave a message, hoping that they will be able to enjoy the long-deprived radiation leave and public holidays, and spend more time with their families!
Then, in the face of the current high misdiagnosis rate of medical images and large gaps, how should we change it? The best answer is artificial intelligence.
Black Technology of Artificial Intelligence——Multilayer Convolutional Neural Network Structure
The application of artificial intelligence in medical imaging is mainly divided into two parts: the first part is image recognition, which has been explained in the previous article; the second part of deep learning is the core part of artificial intelligence application. These two parts are based on the mining and application of data based on medical image big data.
In 2006, Prof. Geoffrey Hinton, a master in the field of neural networks, and his doctoral students published papers in Science and related journals, and for the first time proposed the concept of “depth belief networkâ€. Unlike the traditional training method, the “Deep Belief Network†has a “pre-training†process, which makes it easy to find the value of the neural network in a value close to the optimal solution, and then use “ Fine-tuning technology to optimize the training of the entire network. The use of these two technologies has greatly reduced the time required to train multi-layer neural networks. He gave a new term for multi-layer neural network-related learning methods – “deep learningâ€.
In 2012, Prof. Hinton's research team participated in the ImageNet ILSVRC large-scale image recognition evaluation task organized by Professor Fei-Fei Li of Stanford University. The mission includes 1.2 million high-resolution images and 1000 analogies. Professor Hinton's team used a new black technology multi-layer convolutional neural network structure to dramatically reduce the image recognition error rate from 26.2% to 15.3%. This revolutionary technology allowed neural network deep learning to leap into the medical and industrial fields at an extremely fast rate, which led to the emergence of a series of medical imaging companies that later used the technology.
For example, Enlitic, an internationally renowned medical imaging company, and DeepCare, which has just received a 6 million angel round of financing from Fengrui Capital. Through the accumulation of a large amount of image data and diagnostic data, the neuron network is continuously trained in deep learning, thereby improving the accuracy of doctor diagnosis.
Schematic diagram of Enlitic artificial intelligence assisted medical imaging diagnosis
Taking the malignant tumor detection system developed by Enlitic as an example, it was verified by using the lung cancer related image database "LIDC (Lung Image Database Consortium)" and "NLST (National Lung Screening Trial)", and found that the system developed by the company The accuracy of lung cancer detection is more than 50% higher than that of a radiologist.
All in all, the benefits of artificial intelligence combined with medical imaging are numerous, and patients, radiologists, and hospitals can benefit from the application of artificial intelligence. Artificial intelligence not only helps patients complete health checks more quickly, including X-ray, B-mode ultrasound, and MRI. It can also help video doctors reduce the time of reading, improve efficiency, reduce the probability of misdiagnosis, and assist in diagnosis by suggesting possible side effects.
With the gradual popularization and application of artificial intelligence and medical image big data in the field of medical imaging, the accuracy and large gap problems faced by medical imaging can be solved. The integration of the two will become an important direction for the development of medical imaging.
Source: Arterial Network
Head Massage,Head Massage Spa,Best Head Massager,Relaxing Head Massage
Shenzhen Jie Zhong Lian Investment Co., Ltd. , https://www.szmeizons.com