[ad_1]
Every year around the world, about four in 10,000 children are born with congenital cataract (CC), a rare event according to WHO standards. However, this disease represents 20% of cases of blindness in children. CC can be treated with simple corrective surgery, but in areas of the world where medical facilities are scarce, this treatment is not always an option.
The consequences for children who do not receive adequate medical care can be disastrous: 90% of them will not receive an adequate education, only 50% will survive to adulthood and, of these, 20% only will find gainful employment. But that's where Sinha and Project Prakash come in.
Sinha founded the Prakash project in 2005 as a dual-mission association. "It's a project that stems from the desire to be good scientists, but also to be good Samaritans, to really tackle real-world problems," he said. he told Engadget.
The organization offers eye surgery that changes the lives of many children and young adults with treatable conditions, such as congenital cataracts, that cause blindness. The other mission "seeks to understand how a brain deprived of vision for so many years, if it can learn to acquire visual skills at this stage of life," said Sinha.
It is reasonable to assume that removing cataracts from a child would have the same effect as removing them from an elderly patient – in particular, their sight would be immediately restored – but this is not the case and we do not really know why.
Your average person enjoys a 20/20 vision. That is, they see objects at a distance of 20 feet with the same visual acuity as the rest of the population at the same distance. If you have a near vision or a long-term vision, you just need to wear basic corrective lenses. But the children of Prakash, as Sinha refers to them, often suffer from degraded vision, about 20/100, which is five times more than average.
In addition, their vision can not be corrected with the help of glasses or contacts, because nothing happens physiologically in the eyes or any other element of their optical anatomy. The problem lies in how their brain has developed without visual input during infancy. It is this very problem that ultimately led to Sinha's computer vision, Eureka. But first, a word about infants and their incredible myopia.
Your average baby sees a little less well than Mr. Magoo, in the neighborhood of 20/800 or 40 times worse than an adult. Yet at the age of a few years, their vision has been reduced to 20/20 (or so).
"A big part of the reason [that infants have blurred vision] is that the baby's retina is quite immature, "said Sinha, who also noted that the cones of our eyes, the specialized cells that provide us with high-resolution vision, are much larger in childhood than in the adulthood, reduces the density of these cells in the eye and in turn limits the resolution capabilities of it.As the child grows older, the cones become smaller and more compact, this Although their initially poor vision plays a crucial role in their cognitive development, Sinha theorized that the same concept could be applied to neophyte computer vision systems.
"During the first months, they [children] rely heavily on the movement to analyze the world. So it seems that the visual learning programs being deployed in children's brains seem to recapture some aspects of developmental programs of normally developing infants, "said Sinha.
Sinha's main assumption is that the baby's poor eyesight acts as a set of "training wheels" for the development of his mind. By reducing high-definition details, their formative minds can focus on more fundamental visual and cognitive development without getting lost in details. Evidence suggests that there is a window of opportunity very narrow in the early development of the child for this to happen.
Unfortunately for Prakash children, when they have surgery, this window has been extinguished for a long time. Sinha evokes the case of a Chinese orphan born of congenital cataracts who had been adopted by an American family and brought to the United States for a vision restoration operation at the age of six. While the operation was proceeding wonderfully, the child's adoptive parents began to notice that the child was having trouble making friends.
The child's doctor determined that the problem stemmed from his difficulty in recognizing and memorizing faces – not because of optic problems, but because of the incomplete development of this part of the brain of the child. Child during infancy. "The brain requires a normal visual input related to the face," said Sinha. "And if he is deprived of this entry during this critical window, he will be forever compromised."
In this case, Sinha and his team wondered if artificial intelligence and artificial vision (AI) systems could be formed in the same way as infants. Rather than overwhelming a learning AI with high definition video inputs, could it rather be driven first on fuzzy images before being slowly weaned on more and more inputs? higher resolution. It turns out that the answer is a resounding yes.
Sinha's team formed a deep convolutional neuron (CNN) network on a face database, which had been deliberately blurred to varying degrees. The team trained four of these networks using different treatment regimens: high resolution, high to high blur, blurred to high resolution blur.
<img alt = "asdf" data-caption = "asdf" data-credit = "PNAS" data-credit-link-back = "" data-dam-provider = "" data-local-id = "local-1- 9095171-1558125026851 "data-media-id =" 78e104fe-0abf-4097-ac62-c5e752ae0f48 "data-original-url =" https://s.yimg.com/os/creatr-uploaded-images/2019-05/ 893d0390-78e2-11e9 -b5fb-c202e9d49602 "data-title =" asdf "src =" https://o.aolcdn.com/images/dims?crop=1155%2C1280%2C0&quality=85&format=jpg&resize=1600%2C1773&image_uri= https% 3% 2F% 2Fs.yimg.com% 2Fos% 2Fcreatr-uploaded-images% 2F2019-05% 2F893d0390-78e2-11e9-b5fb-c202e9d49602
The team discovered that when systems were exposed to fuzzy images, its field of reception (RF) expanded. Curiously, "starting with high-resolution images and then introducing fuzzy images at a later stage leads to a significant increase in the size of RF frequencies, big RF." So, no matter when the CV system sees images blurred, only it made. Conversely, this could lead to treatment for Prakash children: artificially blurring their newly restored sight until their brain develops the mechanisms necessary to understand what they see.
"To our pleasant surprise, we found that networks driven with poor quality images actually did better than those that started with higher resolution images," Sinha said. "Having a bad vision at first could force the brain to look at the overall structure of images, rather than focusing on fragments."
In addition, the storage space and bandwidth saved through the use of low resolution images could help reduce training times while reducing the size of training databases that are already too heavy. Sinha's research was recently published in the journal PNAS, although there is no indication yet when this CV training technique could make its way into practical applications. . Although I doubt that it worries many thousands of people who can now see for the first time in their lives.
[ad_2]
Source link