SfN Presidential Special Lecture, Monday 15th November 2010: Learning to See Late in Life - Dr Pawan Sinha, MIT
/"There are two commandments for a scientist: 1) To advance knowledge
2) To help humanity"
...and with this quote began a far-reaching and visionary lecture into how Dr Sinha’s work has managed to marry these two aims simultaneously.
Project Prakash is a humanitarian project with the aim of curing childhood blindness in India. Childhood blindness is associated with a 50% mortality rate by the age of 5 and an employment rate of < 1%. He describes the implementation of school-based screening tests, the more detailed hospital examinations, cataract surgery and the eventual positive outcome for more than 700 children, of 20,000 screened.
However, he also highlights that these operations provide a unique opportunity, one that has arisen only a handful of times over the course of the last millennium, to study the development of vision after the onset of sight in a mature brain. He asks us to begin to redefine our concept of the “critical period” hypothesis, the short period in early post-natal development in which the visual system retains plasticity, after which no further visual development can occur, which is familiar to many of us from our undergraduate teaching, and is included in all modern text books on visual development.
Dr Sinha’s research into the visual function of children that have regained sight demonstrates that the critical period holds true only for features of basic vision – these children do not develop normal visual acuity, oculomotor function (absent nystagmus) or stereopsis. However, amazingly, even in the absence of primitive visual function, higher order visual functions, such as object or face recognition or colour matching can still develop.
The visual changes are also correlated with changes in fMRI activity. Imaging studies demonstrated changes in the “resting state functional connectivity”, studying networks that they called “modules”. Interestingly, they found that at one month post sight-restoring vision, there was a greatly increased area for the neural face response, however at 4 months, this area had become condensed to a smaller area.
An important question to ask is whether this visual recovery actually provides vision that is useful for performance of day-to-day tasks? Having demonstrated behavioural and neural network plasticity at the onset of sight, in a developed brain, the next step was to study the development of useful visual skills. They divided these into 2 categories:
1) Development of “intermodal skill”: inspired by Molyneux’s question posed in 1668: if a blind man can distinguish a two different objects by touch, if he somehow was made to see, having never seen these objects before, would he be able to identify them using vision alone? Dr Sinha finds that in the immediate post-operative period, the newly sighted children are unable to identify objects visually, however after one month they do acquire “touch-to-visual transfer” ability.
2) Development of “intramodal skill”: the integration of features of a visual scene e.g. overlapping shapes – are they seen as one object or many objects? Initially, the children see an overfragmented world – e.g. the patches on a cow or the shadow on a baseball, are interpreted as separate objects. A square overlapping a triangle is interpreted as three objects rather than two. It takes approximately 18 months for the children's fractured world to knit together.
What permits the children to learn to integrate the visual world? Dr Sinha shows us that motion plays an important and dramatic role in helping object recognition – a square overlapping a circle is more readily identified as a square if it is moved over the circle. He concludes that dynamic information processing represents a keystone in the development of higher visual function.
At this point in the lecture, Dr Sinha begins to links into his second area of interest – that of autism. Individuals with autism find it hard to extrapolate the trajectory of a dynamic object. This is studied in more depth using the task of looking at objects through a moving slit. This is tested on the audience – can we identify the picture behind the thin slit that moves across the black power-point slide? Most of the audience seems to be able to do this on the first sweep (iteration). But some visual scenes have key features which give away clues as to the identity of the object. This makes it a bit too easy. A harder task is to use “anorthoscopic words” – and still the audience performs well. This is consistent with the control group in the study, who could identify all objects through a 15 degree slit within 1-2 iterations. In contrast, autistic people need wider slits and many more iterations. Sometimes, even after 10 iterations and wide slits, they still cannot identify the underlying object. What aspects of temporal processing are disrupted in these individuals?
This marks the 3rd adventure of Dr Sinha’s research – the development of a computational system for object recognition using real world visual inputs. This computational system is called Dylan, and its principle information source comes from Dr Sinha’s son, born just 4 years ago! It strikes me that Dr Sinha is good at spotting unique opportunities and grasping them wholeheartedly. In this case, it is emphasised that it took a certain amount of discussion with his wife (notable quotes: “Are you crazy? I am going to report you to the DSS!”) before it was agreed that Baby Sinha should be recruited as a research assistant, to wear a “BabyCam” – a small video camera attached to his baby hat, adjusted to have a resolution to match that of a neonatal visual system, to provide an accurate and dynamic representation of the real visual world of a small baby. This “real world” visual information is then fed into Dylan. The first computational step is the extraction of low level salience features, including colour, luminence, motion and novelty. These features are then further processed into a composite salience signal. And remarkably, after only 8 minutes of real visual input, Dylan demonstrates oriented behaviour towards areas of “peak interest” in the visual world – it localises to the centre of faces. The fascinating conclusion is that a relatively simple computational system can very rapidly develop a rudimentary “face concept” (something that till now has always been considered a very complex higher visual function) using simple real world dynamic visual information.
Towards the end of the lecture, having swept through so many far-reaching goals, Dr Sinha reminds us that we should never cease to think about what more we can do for knowledge and for society. As Einstein said “Those who have the privilege to know have the duty to act and in that action are the seeds of new knowledge”. Not satisfied with curing blindness, discovering important fundamental principles about development of higher visual function and developing artificial intelligence computational systems for face recognition, Dr Sinha feels that Project Prakash needs to reach out to many more children and emphasises that we are still only at the very small tip the iceberg of childhood blindness. He has achieved two of his goals: those of improving health and developing new knowledge. The third responsibility and goal is that of providing education. He describes to us his plans for the “Prakash Centre for Children” – a centre to combine a 50-bed ophthalmology hospital, a vision research centre and a school.
There ends an inspiring lecture full of not only very good science, but the reminder that even as scientists we have the opportunity and responsibility to help society in as many ways as our imagination permits.
