The product is mainly for the comprehensive disease-assisted diagnosis system, and the voxel technology uses AI to empower existing clinical workflows.

Recently, the Arterial Network held a second 2018 primary medical forum based on the “new structure” in Hangzhou. Ding Xiaowei, the founder of Voxel Technology, gave a wonderful speech with "AI-Doctor's Right Arm" and accepted an exclusive interview with the Arterial Network. The following is an interview:

Ding Xiaowei: Assistant Professor of Computer Science, UCLA, Ph.D. (Academician of the Oscar Award, Terzopoulos): After graduating from UCLA to UCLA (Computer, Minor Applied Mathematics), it took less than three years to break the UCLA computer science degree. Former Cedars-Sinai Medical Center Researcher, Artificial Intelligence Medical Department, Daniel S., the best doctor of the United States, founder of Cardiac Nuclear Medicine Berman, famous expert in magnetic resonance imaging, Debiao Li.

From a medical professional point of view, in order to better empower existing clinical workflows, Voxel Technology does not use disease protocols to locate product types. Many products in the industry are developed for single disease or single lesion types, such as the diabetes network auxiliary diagnosis system, lung nodule auxiliary diagnosis system and so on. Voxel Technology's products are based on image protocols to locate products. For example, chest CT solutions, fundus image solutions, chest X-ray solutions.

Taking the chest CT solution as an example, when examining a lung image, the radiologist should pay attention to whether the screener has emphysema, bronchiectasis, pneumonia, mediastinal lesions, chest wall lesions, and scan of the upper abdominal lesion. Wait for chest disease. First-line radiologists hope that the future AI system can conduct a comprehensive examination of chest diseases like a doctor, not just a lung nodule screening, and finally a preliminary inspection report.

The whole disease assisted diagnosis system is suitable for initial diagnosis

Ding Xiaowei told the arterial network that the incidence of lung cancer is very high, but this high incidence is compared with other cancers, but the incidence is much lower than other common diseases.

If the patient is diagnosed for the first time, he or she usually only knows the symptoms and high-risk factors. After the clinical department, the doctor will recommend the patient to take a film according to the situation. Before the radiologist gets the film, he will not know in advance where there is a problem with the radiological lung image, nor will he assume which disease the screener has.

So according to the natural logic, this paragraph starts with the image protocol. The imaging protocol refers to chest CT, fundus photography, coronary CT, etc. performed by the patient.

For the returning patient during the treatment, the doctor knows the specific condition of the patient, and the doctor's request is to monitor the patient's treatment effect. This is not a category of diagnosis, but a category of management observation. Therefore, voxel technology adds pre- and post-contrast to the corresponding lesions in the product.

For those patients who have no initial diagnosis and return to confirm the disease, such a condition can be presupposed to assume a certain disease, because the doctor has previously suspected that it is a certain disease, this time just to confirm. In response to this situation, voxel technology's single disease products and all disease products can be covered.

Ding Xiaowei believes that the single disease auxiliary diagnosis system is used in the follow-up examination, and the comprehensive diagnosis system for the whole disease is used in the initial diagnosis to meet the doctor's work requirements.

The whole disease auxiliary diagnostic system has high data requirements, but it is convenient for subsequent grinding.

Ding Xiaowei believes that the medical artificial intelligence industry will not suddenly rise to a particularly powerful team, which takes time to accumulate. Before the establishment of Voxel Technology, Ding Xiaowei engaged in research work with doctors almost every day during his research on medical image artificial intelligence. He had a certain understanding of the diagnosis and treatment of diseases. Although he could not see a doctor and treat incurable diseases, the basics were The medical image can still be understood. Many researchers who have been in the medical industry for a long time will slowly understand some medical treatment.

Based on his own understanding of medical imaging, Ding Xiaowei's team established the company's main products in the comprehensive disease-assisted diagnosis system. Compared with the single-path auxiliary diagnosis system, this type of positioning has a high number of labels, and the data labeling and quality requirements are very high. However, this development idea saves a lot of energy for the subsequent product polishing.

Because machine learning, especially deep learning, has many parameter labels, it needs to be polished in the case of strong supervision. Therefore, the label is too simplified, but it is not conducive to learning, so the system can screen enough diseases.

Take the detection of pulmonary nodules as an example. If the system can only screen for pulmonary nodules, the positive (exceptional) part is the pulmonary nodules themselves, except for the lung nodules themselves, which are all negative (normal). This is a very difficult thing for the model, because there are too many types of negatives, such as blanks, blood vessels, other local lesions, and so on.

This creates confusion for the system, why it is negative, and why the looks are so different. The idea of ​​voxel technology is to classify the negative areas according to the type of lesion or the type of tissue, so the system will be very clear. Although it is very troublesome when designing the model, the subsequent adjustment of parameters and supervision of these labels will be simple. a lot of.

The idea of ​​fundus images is the same. If you only detect sugar net disease in the diabetic population, other abnormalities can easily be classified as false positives. Instead of telling the system directly what the various types of exceptions are.

Ding Xiaowei said that the product is positioned in the comprehensive disease-assisted diagnosis system, which not only meets the doctor's workflow, but also saves energy in product development, which can make the system's performance reach the extreme.

Ding Xiaowei also stressed that this training method is also difficult. After the number of diseases is large, the balance between the diseases is difficult to do. Some diseases are common and some diseases are rare. If you use unbalanced data to train during training, the effect is not good, and rare diseases will be ignored.

In addition, the relationship between diseases is difficult to edit in a model. For example, some are exclusive, some diseases coexist, some are intersections, some are hierarchical, some are not related, so balance and handle each The relationship between diseases is more difficult. The classification of comorbidities also requires detailed instructions from the doctor during the development process in order to quality control the markers.

Voxel technology part of the mature solution

Based on the above product positioning ideas, Voxel Technology has developed a chest CT scan solution, secondary diagnosis of lung cancer, CT coronary angiography solution, and retinal disease solutions.

Chest CT solution

Voxel Technology is developing a chest non-enhanced CT solution that analyzes individual diseases and generates natural language reports with quantitative analysis of abnormal sites. In order to be a standard description, there are many models behind the voxel technology products to run. It is not simple to report the anatomical location of the lesion, the description of the lesion, the nature of the lesion, the quantitative information and the clinically meaningful normal findings. The engineering system can be completed.

In order to better diagnose lung cancer, voxel technology solves this problem from both micro and macro perspectives. Cancer can be analyzed from a molecular scale, and it also requires focal problems after protein synthesis from a macro perspective.

From a macro perspective, voxel technology describes the shape of this lesion from the order of millimeters. From a molecular point of view, the voxel technology and collaborative liquid biopsy research team conducted ctDNA detection by non-invasive blood sampling, and placed the two images of image signage and gene sequencing in the same context to give a definition of lung cancer.

On CT, the sensitivity of malignant tumors is relatively high, the specificity is not high, and ctDNA has high specificity. Therefore, the combination of the characteristics of these two tests can give patients an early diagnosis of secondary diagnosis.

Cardiac coronary angiography CT solution

Then there is coronary heart disease, the first human killer. In China, if the elderly in the family feel chest pain and chest tightness, they will go to the hospital for a common examination - coronary CT angiography. Through this CT, you can see the degree of stenosis of the three coronary arteries and the cause of the stenosis.

Voxel Technology Autoplaque provides, in addition to the functionality of a post-processing workstation for common coronary CTA, it also provides quantitative analysis beyond the requirements of clinical procedures, such as the morphology and composition of some plaques in the heart lumen. For cardiovascular disease, voxel technology can tell patients the prediction of cardiovascular invasive examination in advance through non-invasive examination, reminding him what kind of examination to do next, and what time interval is followed.

In order to accurately predict the sudden onset of heart disease, Voxel Technology quantifies the fat of the heart, several narrow plaques in the blood vessels, etc. on the image of a patient's heart, and obtains quantitative indicators, which are combined with the patient's clinical information, such as biochemistry. Indicators, blood indicators, medical history, family history, etc., predict which person has the highest probability of heart attack in the next five years.

In order to make the model more accurate, voxel technology in 20 hospitals, followed up with 50,000 patients, reasoned by these real data.

Chest X-ray screening solution

Chest X-rays cannot diagnose all chest and lung diseases, but can detect some abnormalities. Therefore, it is quite important for a healthy medical examination patient to be admitted to the hospital for further diagnosis.

This positioning is very reasonable if it can be used for patient triage and referrals. So voxel technology has done all the initial screening of early lesion types in this field.

Retina full disease solution

In the field of ophthalmology, voxel technology not only screens for sugar net diseases, but also does a retina full-thickness solution, and also gives a diagnostic analysis report. At present, this product of Voxel Technology covers most of the eye diseases, and it is also very good. It has been recognized by the NIH project in the United States. The selected Voxel technology products are tested in the entire Los Angeles County. In China, Veterinary Science and Technology is also conducting clinical trials in three more authoritative hospital ophthalmology departments in Beijing.

Children Face Mask

Children Face Mask,Kids Medical Disposable Face Mask,Kids 3 Ply Disposable Face Mask,Kids Nonwoven Disposable Face Mask

KUTA TECHNOLOGY INDUSTRY CO.,LIMITED , https://www.kutasureblue.com