Google has partnered with the Max Planck Institute of Neurobiology in Germany to develop high resolution imaging of brain and neural connections.
Connectomics is the research field that imaged the brain and nerve connections. It is said to be connectedness. If we can image brain synaptic connection information and neural connection information at high resolution, we can see at a glance the difference between a healthy person and a non-healthy person. It is expected that effective treatment will be possible.
According to Google, the technology developed can image 1mm³ tissue and generate data over 1,000TB. It is possible to create a detailed connection map.
The technology announced by Google and the Max Planck Institute for Neurobiology uses Recurrent Neural Networks (RNN), known as one of the machine learning algorithms used for handwriting recognition and speech recognition. Of course, similar technology has already been announced by Intel in March with Image Technology using deep-run in collaboration with MIT. However, according to Google, the technology can be rendered with a precision that is more than ten times higher than that announced by Intel at the time. Google has released the source code that applies the technology to the tensor flow for free on its hub.
Google is constantly trying to integrate artificial intelligence into the medical field. In March, he collaborated with Duke University’s research team to develop a system to automate protein crystallization experiments using artificial intelligence. The biological function of a protein depends on its molecular structure. Molecular structure analysis is important to develop effective new drugs. Protein molecular structure analysis requires protein crystallization, but protein crystallization still has many unknowns.
In protein crystallization experiments, most of the preparations and processes have been automated, but the results of crystallization experiments still have to be confirmed by human microscopy. In this case, you have to depend on human observation and experience, and you may miss out on the crystallization.
Duke University said that it could reduce the risk by automatically recognizing the result of protein crystallization by artificial intelligence through Machine Recognition of Crystallization Outcomes (MARCO) and crystallization outcome machine recognition, and cooperate with Google on more than 500,000 protein crystallization experiment data. I asked. Google has developed a system that can automatically classify crystallization results through neural networks.
This allows the artificial intelligence to visually identify protein crystals with high precision by learning the vast amount of data collected by MARCO. According to the researchers, the precision of protein crystal identification is as high as 95% when using artificial intelligence, compared to 85% for humans. By automatically discovery or classification of protein crystals, artificial intelligence can almost automate protein crystallization experiments. This will be beneficial to protein structure analysis and greatly reduce the time and cost of new drug research. Google is also releasing this model on its flagship hub.
It is not in the New Testament, but the announcement in April can help in a similar vein. We have developed an artificial intelligence technology that analyzes optical microscope images and detects cancer in real time. When a patient is cancerous, the doctor observes the body tissue taken from the patient with an optical microscope to determine the presence of cancer tissue. What Google has developed is to allow the artificial intelligence to detect what looks like microscopic patterns and look at cancer in real time. To eliminate the time and fatigue that a doctor takes when diagnosing cancer. In order to find a cancer, it is necessary to examine all the cells that have been collected and sometimes it is difficult to detect cancer. This problem is solved by combining optical microscope and artificial intelligence technology.
The artificial intelligence detects the image of the optical microscope and finds an organization similar to the learned mass pattern. Artificial intelligence finds an organization and marks it on the border so that the doctor can easily judge it. The artificial intelligence synchronizes the frame position and the image with feedback at a rate of 10 frames per second so that no error occurs between the image and the projected image in the detected frame. By providing such a diagnostic aid in real time, artificial intelligence will improve cancer diagnosis accuracy and reduce time. The developed algorithm will operate at 4 ~ 40x magnification and will be used for diagnosis of major cancer.
Google’s attempt to reduce medical waste in modern healthcare through artificial intelligence algorithms is an effort to simplify it. Google has also developed algorithms for estimating patient mortality probability before entering the medical field. In fact, a late-onset breast cancer patient underwent a radiological examination at a hospital and was analyzed by a computer in the hospital. The probability of death of the patient was estimated at 9.3%, but the risk of death doubled to 19.9% It is estimated. A few days later, the patient died.
Google announced a new algorithm in May that can predict the probability of patient death. By automatically learning the data, you can pull up the performance yourself and predict how long the patient will stay in the hospital and how likely it will be for re-admission. This algorithm predicts what will happen to the patient much faster and more accurately than the existing technology.
Of course, existing hospitals have tried to improve the quality of medical services based on their health records and patient data, but collecting and analyzing medical data requires considerable time and money. Most of the software that has been used for medical purposes is manual coding. In contrast, Google’s algorithm takes an approach to learning how to analyze data automatically.
If Google is successful in trying to combine artificial intelligence with healthcare, then it is likely that Google will have the ability to monopolize data in healthcare. Of course, algorithms can save time and money, and as a result, you can expect to deliver value through it. However, it can be said that medical data is not easily free from related laws and regulations (at least for Google). The combination of medical and artificial intelligence has given me the task of how to create new value and how to combine gold (data) and regulations around it in new markets.