In two new studies, scientists from the University of California at Los Angeles (UCLA) used neural networks to reconstruct holograms. Both works not only demonstrate the level of development of holographic technologies, but also promise to open these technologies to the door to medicine, where they can make a real revolution.
In the first study, the results of which were described in the journal Light: Science & Applications, scientists used in-depth training techniques to create images of biological samples: blood, Pap smears, and also some other samples of biological tissues.
The research proved that the use of neural networks significantly accelerates and simplifies the process of creating holographic images, in comparison with the more traditional methods of creating similar images that require for the reconstruction of the object under study preliminary physical calculations and manual computer data input.
In the second study, a team of specialists used their deep training framework to improve the resolution and quality of microscopic images that help doctors determine the smallest, barely noticeable abnormalities in large blood and cell tissue samples.
One of the problems with current holographic methods is that during the rendering process, holograms can lose some information, which in turn can lead to the appearance of “artifacts” in the final image. Sometimes these losses are very significant. For example, black dots can appear on the image, which doctors mistakenly take as the growth of cancer cells. Such artifacts are often found in radiological scanning, especially if the patient starts moving when the scanner does its job.
The system of in-depth training of the University of California demonstrated its effectiveness in solving this problem. Once the system is properly trained, the neural network can easily separate the spatial features of the present image from any external interference (in which the role of light often appears).
Multilayered artificial neural networks allow deep learning algorithms to analyze data in an autonomous mode. The technology has already demonstrated its effectiveness in the example of translation of speech from one language to another in real time, video capture of images, as well as in many other tasks that a person had to cope with before, which, by the way, loses algorithms in speed of execution these tasks.
Since machine learning systems have acquired the ability to sort and analyze vast amounts of information much faster than humans, it is not surprising that a wide range of areas, including medicine, begin to show interest in these technologies. Algorithms already find their application, for example, in diagnostic radiology, where they demonstrate their effectiveness in reading X-ray images, as well as searching for cancer cells that could have been missed by doctors during scanning.
Holographic technologies are no longer treated as they were before, when they were considered more of an object of science fiction than a practical tool. Now scientists are confident in the prospects of this direction.
Methods of in-depth training, in turn, can help in this direction, says Aidogan Ozkan, a leading researcher. In his opinion, these technologies will open new opportunities for visualization. In a press release from the University of California, Ozkan noted that such technologies can even lead to the development of entirely new coherent image processing systems. The scientist believes that the developments of UCLA can be used to further improve the technology and incorporate into it the support of other parts of the electromagnetic spectrum, for example, X-ray and optical radiation.
If the future awaits us, which we could see in science fiction for the last 40-50 years, then the holograms in it will play definitely not the last role. UCLA research in this direction, in turn, is not just trying to support this fantastic technology, they offer real environments for its application.