August 30-31, 2018 China (Shanghai) International Artificial Intelligence Exhibition and OFweek (2nd) hosted by China High-tech Industry Portal OFweek, the High-tech Association, OFweek Artificial Intelligence Network and OFweek Medical Technology Network The Artificial Intelligence Industry Conference was successfully held at the Shanghai International Sourcing Exhibition Center.

The conference brought together thousands of elites from the world's leading enterprises in the field of artificial intelligence, senior industry experts, expert analysis institutions, etc., to analyze the industry dynamics with a new perspective, and to interpret the most noteworthy academic and R&D progress in the field of artificial intelligence at home and abroad this year. .

In the August 31 sub-forum: AI+ Medical Special Session, special researcher and PhD supervisor of the School of Biomedical Engineering, Shanghai Jiaotong University, brought a keynote speech on “Medical Image Information System: Auxiliary Detection, Diagnosis and Exploration” from the audience. Intelligent technology has been discussed in detail in the development of medical fields, artificial intelligence and medical imaging information systems.

Special Research Fellow, School of Biomedical Engineering, Shanghai Jiaotong University, Ph.D.

Qian Xiaohua explained the current application and development prospects of artificial intelligence in the medical field from various angles. Qian Xiaohua believes that the development of artificial intelligence technology has greatly promoted the progress of the medical industry, but medical artificial intelligence is impossible to replace doctors in essence and can only be applied in the field of medical assistance.

In the speech, Qian Xiaohua also mentioned the past research with American doctors on understanding the progress of true and false glioma surgery. In his view, past doctors believe that the development of artificial intelligence technology is not yet mature, and artificial intelligence is a black box for people, and its incomprehensible knowledge and characteristics hinder the progress of people's medical technology. Therefore, Qian Xiaohua's team conducted a feature interpretability study on the deep learning link in the project, and analyzed the abstract features in deep learning into the knowledge and characteristics that doctors can understand. The doctor judged the evaluation labor through the explained knowledge and characteristics. Intelligent diagnostic results to promote the development of artificial intelligence technology in the medical field.

The following is the live speech of Mr. Qian Xiaohua. The OFweek Medical Technology Network has compiled and edited without changing the original intention:

Next, I will give you a report entitled "Medical Image Information System: Auxiliary Detection, Diagnosis and Exploration." First of all, I think that medical artificial intelligence is impossible to replace doctors and can only be applied in the auxiliary field. I simply introduced myself. I joined the Shanghai Jiaotong University School of Biomedical Engineering last year and set up my own medical imaging processing team. At present, there are seven graduate students. The cooperative hospitals include Ruijin Hospital and Huashan Hospital. , Chest Hospital and Beijing Chaoyang Hospital. Next, I will start today's speech topic from the following aspects.

First, it is based on an early detection system for non-enhanced ischemic stroke, and the goal is to provide early warning and quantitative diagnosis of ischemic stroke. A variety of modal nuclear magnetic resonances and various methods can be used in scientific research or in later diagnosis. However, in the early examination process, ordinary CT is used. For example, if a patient goes to the emergency department, the MRI can not be performed without diagnosis. This work is based on the needs in the clinical process. This work has been carried out for 7 or 8 years. The current work effect is that when the suspected area and the surrounding area are also more than 5 CT values, the accuracy of the system is more than 80%.

Second, introduce the computer-assisted surgery system. Taking infants with craniocerebral dystrophy as an example, the symptoms of infants with craniocerebral dystrophy are that the forehead of the child is convex like "Shou Xing". This is actually the early closure of the child's cerebral sulcus, which will cause the intracranial pressure to rise after closing. High, affecting children's intellectual development. The treatment is straightforward, cutting the fused seam directly and then inserting the two springs. How much spring does this spring need? It is up to the doctor's experience to decide, and this experience comes from the previous mouse experiment. The first doctor in the United States who invented this operation had worked with us. He is very eager to solve this problem, and he hopes to customize the personalized spring according to the actual situation of each child.

As you can see, this is a typical image processing, machine learning plus preferred analysis of the surgical planning system. The difficulty and key point of the whole system lies in the segmentation and extraction and characterization of the brain crust, which is the biggest difficulty of this project. The human eye is very easy to see the brain, but the computer does not understand, because the brain seam does not exist in nature. The current deep learning still can't solve this problem.

This is a very interesting question. Deep learning often requires a large sample size. Many intractable diseases have a small sample size, and you urgently need to objectively make this work, so that more people can learn the system, then you need traditional methods or other ways to complete the work. We have come up with a solution: no one has ever done this based on hemispherical projection. How to project and seem to have nothing to do with segmentation and extraction in a closed area is the biggest difficulty. Our solution was supported by the US Department of Health's $2.5 million research fund.

The third job is about understanding the progress of true and false glioma surgery. Gliomas are a malignant tumor with a survival of only 18 months. The key problem is that the condition is prone to recurrence. Its treatment is first post-operative radiotherapy and chemotherapy, and radiotherapy and chemotherapy will bring a side effect or false recurrence. False recurrence The gray-scale enhancement region shown on NMR mimics the pattern of true recurrence. Even a professional doctor can't judge whether the patient is really recurring or having a recurrence by film. One of the current clinical criteria is follow-up, which usually takes more than six months to determine the patient's condition based on its morphological changes. Gliomas live for a total of 18 months. If you spend more than six months to diagnose, it will seriously affect the patient's life.

So American doctors have found us and hope to shorten the diagnosis time. Around this demand, we carry out the following three aspects of work. The first is to develop a computer-aided diagnosis system based on image technology and artificial intelligence technology to improve accuracy and diagnostic efficiency; the second is to determine biomarkers through image and gene combination, so that Biomarkers are used to predict the patient's condition as early as possible after the surgery is completed; the third is to explore a new staging method. In the past few years, we have independently developed a dictionary learning system for segmentation, and its accuracy is between 80% and 85%. The key to this system is not the system itself, but the comprehensibility of dictionary learning. At that time, doctors believed that the advanced technology of artificial intelligence was nothing more than a black box for people. It would not be used without understanding its characteristics. Therefore, we have carried out research on feature interpretability problems. In the past two years, we have done this diagnosis of true and false recurrence through deep learning. The accuracy of the diagnostic system shown in the figure has increased from over 80% to over 90%. But we still don't trust this. Because the nature and internal logic of this accuracy is not consistent with the doctor's logic. So now the focus of our work is to uncover the artificial intelligence black box and see if these abstract features really meet the doctor's standards to determine whether the system is reliable.

The second job is to identify a biomarker by means of image and gene binding. Because different treatments are used to treat different patients, different treatment outcomes are obtained. We believe that this is caused by different patients' genomes. Therefore, it is necessary to combine the image technology and gene technology to mark the genes. After the operation is completed, the genetic test can be used to predict the development trend of the patient's condition.

In the third work, we conducted a statistical analysis of the clinical records of true and false recurrence of a hospital glioma. It can be found that there is no significant difference in survival time between the two types of patients with true and false recurrence after different treatments. This illustrates two issues. The first point indicates that a clinically spent six months of diagnosis is a delay in treatment progress. The second point indicates that the current standard of clinical diagnosis is based on image judgment, and the appearance of the image does not necessarily reflect the essential difference between the two types of patients. Therefore, we propose to combine image technology and gene technology to construct a two-year survival risk prediction model. According to the two-year survival value generated by the predictive model, the patient is divided into high-low risk patients. This will allow you to know the direction of further clinical diagnostics.

Next, I will briefly introduce other work. The first is an in-depth study of in-depth learning models. A well-modulated deep learning system can provide good diagnostic prediction and high accuracy, but it does not necessarily match the intrinsic logic to the clinical standard.

As shown in the figure, the ROC value can reach 99% in the whole process, but we look back through the feature and find that the areas with the largest contribution of these abstract features are basically not in the mainstream part, but the noise characteristics contribute the accuracy. Therefore, the deep learning system and the deep learning system are not completely consistent. The reliability mainly comes from the comprehensibility of the model. The abstract features in deep learning are interpreted into the knowledge and characteristics that doctors can understand. The doctor explains the knowledge and Feature judgment evaluates the diagnosis results of black box, which is the focus of our work. In addition, we have developed a decision-making system for brain metastases based on imaging and genomics, and an early diagnosis/prediction system for pancreatic cancer. The above is the speech that I brought today, thank you!

Automatic Massage Foot Bath Machine

The automatic massage foot bath machine is a device that uses water and massage rollers to provide a relaxing foot massage. It usually has a basin filled with water and has built-in massage rollers that move and massage the feet. The machine may also have other functions, such as heat therapy, air bubbles, and vibration. Users place their feet in the basin and the machine provides a soothing massage that helps relieve tension and improve foot circulation. Some automatic foot massage machines also come with removable attachments for additional massage options.

Automatic Massage Foot Bath Machine,Foot Spa Massageer,Heated Foot Spa,Foot Bath Massage Basin

Huaian Mimir Electric Appliance Co., LTD , https://www.mmfootspa.com