Technology Innovation: Finance, Social, Healthcare

As a means of conducting transactions without requiring an intermediary, blockchains have many potential applications such as in peer-to-peer electricity trading. In current blockchains, an electronic signature is affixed to transaction data to guarantee its trustworthiness. If the private key by which a user generates electronic signatures is compromised or lost, however, they may lose the assets they have traded on the blockchain or may be a victim of fraudulent transactions made by someone using their identity.

In response, Hitachi has developed a technique for reducing blockchain risks that uses its proprietary public biometrics infrastructure (PBI) for generating private keys from biometric information. Because the technique allows users to generate their private key from biometric information whenever they need it, they no longer need to store the key on an authentication device and are at much lower risk of having their key lost or stolen.

Hitachi intends to proceed with the practical implementation of this technique, including by engaging in demonstration projects with corporate customers.

The growing use of financial transactions in digital services, such as payments and the use of vouchers, has created a demand for a reliable means of personal identification (ID) similar to that used with bank tellers in the past.

In response, Hitachi has developed a technique for generating digital IDs that utilizes its proprietary PBI technology and can reliably identify a person using biometric information acquired from a conventional mobile device such as a smartphone. Accompanying this, Hitachi has also developed a digital ID platform that enables the digital ID to be shared and reused not only by institutions with which the user has registered but also by local government, financial institutions, retailers, and various other participating organizations. The risk of identity theft or data tampering is reduced by using a blockchain, a technology that is very resistant to data falsification, meaning the platform can be used to offer new services on which organizations can work together with confidence.

Hitachi intends to expedite the commercialization of the platform by engaging in collaborative creation with corporate customers.

The prediction of rarely occurring events is an important part of tasks such as credit decisions regarding bad debt or assessing the validity of transactions, for example. While the conventional practice with deep learning is to utilize learning from actual data to reduce the error of prediction formulae, the lack of such data for rarely occurring events makes this difficult. Another issue is that attempting to improve the accuracy of predictive models tends to make the formulae more complex, making it more difficult to explain the reasoning behind their predictions.

Now, Hitachi has developed an artificial intelligence (AI) that can predict rarely occurring events with a high level of accuracy and describe the basis for its predictions. This involved the development of signal noise learning, a technique that is trained not only on data acquired under normal conditions, but also on how not to be influenced by biased or outlier data. Testing in which this was applied to loan data and used to make bad debt predictions found a 43% higher accuracy than when using conventional deep learning.

Advances in the Internet of Things (IoT) are providing railway operators with the ability to collect a variety of data from both onboard and off-board equipment, bringing with it rising expectations for using this data in applications like condition-based maintenance (CBM) and predictive maintenance (PdM) that can boost maintenance efficiency without compromising safety. While modeling the behavior of machinery and other equipment to identify potential faults and consider countermeasures plays an important part in implementing CBM and PdM, a problem with using conventional machine learning for modeling is that its operation is difficult for humans to interpret, making it hard to deploy in practice.

In response, Hitachi has developed a cause-finding technique that can automatically identify cause and effect relationships in data. The technique involves creating a hierarchical model of the relationships between data and automatically calculating the degree of contribution of lower-order variables (1), thereby making it easy to trace the causes of events. Moreover, in the case of sets of variables for which it is difficult to untangle cause from effect, the technique can be used to revise models in a way that can convince those who will use them in practice by using automatic grouping to create alternative variable candidates (2).

Hitachi intends to deploy the technique in railway maintenance by implementing it as a support system for boosting maintenance efficiency without compromising safety.

Hitachi, in collaboration with Partners HealthCare, developed AI technology that accurately predicts heart failure patient readmissions.

In the USA, the Hospital Readmission Reduction Program has been in action since 2012. To reduce readmissions, Partners HealthCare implemented a remote monitoring program for monitoring daily symptoms, blood pressure, and weight, plus just-in-time feedback about measurements from the clinical team for patients with heart failure. Even though the program has been largely successful, there is still a need to better stratify patients based on risk to achieve greater reductions in readmissions.

To address this challenge, Hitachi developed a new deep learning algorithm (recently known as explainable AI), that sorts about 3,500 risk factors in order of weighted importance and leveraged this information successfully to achieve an accuracy of area under the curve of 0.71. By using this AI tool, Hitachi and Partners HealthCare are now exploring new opportunities to achieve better post-discharge care coordination support by making actionable information available to multidisciplinary care teams.

To shorten the time taken for ultrasound scans looking for disorders of the cardiovascular system, Hitachi has launched a three-dimensional (3D) matrix probe^*1 and the LISENDO 880^*2, its high-end scanner model.

The newly developed matrix probe contains a large number of tiny ultrasonic transducers arranged in a matrix (grid) as well as a small integrated circuit that processes the ultrasound signals at high speed. It can instantaneously acquire precise 3D data for the entire heart simply by placing it on the subject's body.

The LISENDO 880, meanwhile, uses a high-speed image processing engine to generate high-quality 3D volume images from the data acquired by the matrix probe. It is also equipped with an application that utilizes machine learning to automate the various measurement and analysis tasks that are needed for cardiovascular diagnosis. These help to significantly shorten the time required to perform an ultrasound scan.

By expanding the scope of application of 3D imaging and making greater use of AI, Hitachi intends to help deliver customer value in ways that impose less stress on patients, improve diagnostic quality, and increase hospital scanning throughput.

*1

Product name: MXS1 probe (Medical device certification number: 228ABBZX00097000)

*2

Product name: ALOKA LISENDO 880 (Medical device certification number: 228ABBZX00092000)

Automatic analyzers are powerful instruments for blood testing at hospitals and, when capable of measuring a wider range of concentrations, can help improve operational efficiency by allowing a wide variety of tests to be done using the same analyzer. Hitachi automatic analyzer 3500 launched by Hitachi High-Technologies Corporation in 2017 was the first in the world to incorporate a scattered light photometer, achieving its multi-function capabilities by utilizing this enhanced sensitivity to measure a wide variety of tests.

When an automatic analyzer is used for immunoassay testing, latex beads sensitized with antibodies are used as a reagent and mixed with the blood sample, with the concentration of the different components being measured from the change in turbidity due to agglutination (clumping) in the reaction liquid. The conventional method is to perform this measurement by using an absorption photometer to detect transmitted light.

For the new technique, Hitachi has developed a scattered light photometer that can measure lower concentrations than previously possible by selecting a light source wavelength and an angle for scattered light measurement that is appropriate for the latex beads. This technique allows a wide variety of tests by using the scattered light photometer to measure low concentrations and the conventional absorption photometer for high concentrations.