Picture Perfect

We watched Photograph 51 last week. A play written by Anna Zeigler, it is based on the role that an x-ray diffraction image probably played in unraveling the structure of the deoxyribonucleic acid (DNA). In science just as in all other pursuits in life, there is a journey and a destination. My own humble life in science has taught me that the path we travel is influenced to a considerable degree on whether we are seeking the truth or a victory.

I read the biography of Rosalind Franklin when I was about forty. Up until then, I knew that James Watson and Francis Crick had discovered the DNA. I had not heard of Maurice Wilkins or Raymond Gosling. Science lessons are supposed to be objective and only state facts; not provide sociological history. May be I was too young when I gave up biology at the end of 8th class …?

Dr. Rosalind Franklin moved from Paris to London in 1951. She had expected to be given charge of an independent laboratory where she could pursue her research on x-ray diffraction in King’s College; instead she found herself mandated to assist Dr. Maurice Wilkins in his search for the structure of DNA. The disappointment caused by this arrangement and the many assumptions that Dr. Wilkins unilaterally made about their mode of collaboration got the team dynamics off to a dismal start from which it never recovered and eventually contributed to the drama that unfolded.   At least two of those assumptions continue to be endemic even today in research laboratories. Dr. Wilkins transferred his PhD student Raymond Gosling to Dr. Franklin without consulting either of them. He saw Rosalind as the experimentalist who would generate the x-ray photographs while he, Maurice, the theoretician would apply his mind to the process of discovery. That Rosalind was a woman, a Jew, single and singularly determined to follow a career in science was in an institution where women were not allowed into the senior common room where lunch was served was not inconsistent with the dominant moorings of those times.

Unlike Rosalind and Maurice, the chemistry of the collaboration between Watson and Crick was magical. James, trained as a Zoologist and, Francis, a Molecular Biologist were both drawn to genetics and the DNA. Much has been written about their very distinct personalities and styles but that seems to have only fuelled a synthesis of ideas and lent momentum to their desire to be first in the race for a Nobel. Watson was adventurous and inventive. Having obtained a doctorate in the United States by the age of 22, he had figured out that the fastest way to get to the goal post was to propose, test and dispose hypotheses rapidly until you got to the one that fell in place. He and Crick had a strong intuition that x-ray diffraction would provide a clue to the structure. That was his reason to choose to work at Cambridge, to be close to Lawrence Bragg [i], the then Director of Cavendish. Watson had no fear of intermediate failure, no lack of self confidence, no qualms in looking around the corners or walking across to others like Maurice Wilkins to hunt for new information or data to bolster the models that he and Crick were building. Rosalind was a Physical Chemist and her approach was to arrive at a solution through perfecting the quality of data that would allow for calculations, reliable analysis and interpretation. Maurice was a Bio Physicist; while he admired Rosalind’s attention to detail, he wished she would relent and be more open to injecting a model based thought process into her methodology but by then the door to a dialogue had already closed. All through 1951-52, Rosalind worked tirelessly to improve the x-ray system in the lab so as to obtain pictures with sharper clarity and detail. It is this strive for perfection that led to the Photograph 51 that was taken by Gosling and the one that Maurice Wilkins shared with James Watson in a moment of despair as he frustrated over having to work with an uncompromising Rosalind. He did so without Rosalind’s knowledge. The x-ray diffraction pattern in the photograph helped Watson and Crick to work out the details of the double helix such as the placement of the phosphates (whether they were inside or outside), the pairing of the bases and the sizes of the molecules involved and present these to the scientific community in early 1953.

The Nobel was finally awarded in 1962 to Watson, Crick and Wilkins. Ovarian cancer possibly caused by over exposure to x-ray claimed Rosalind in 1958. She may have independently arrived at the structure of DNA from a detailed study of Photograph 51 by herself. But, unlike Watson, she did not have one or more models in mind. So, while she recognized the superior quality of the photo, it did not at the first glance provide her the “Eureka moment”.

A stellar performance by the cast depicting Rosalind Franklin, Maurice Wilkins, Raymond Gosling, Francis Crick, James Watson and Donald Casper in the 90-minutes show of Photograph 51 at the Jagriti Theatre [ii] brought back the emotions that had swept over me when I had first read the biography. There was also the undeniable pleasure of seeing our good friend Sanjeev Gadre on stage in the role of Maurice Wilkins who I have always thought of as the most complex and nuanced character in the DNA story.

I would go on to say that Maurice Wilkins abound in Research Labs even today. They mean well, they are decent human beings but they are confused and often uncomfortable with capable and determined women colleagues. The Watsons are smart and brash, interested mostly in winning the race and will do anything to be the first to get there. They are relatively rare but know a Crick and a Wilkins when they see one. The Cricks are hard-working, brilliant and successful – but they are postponing a future they so desire, a future they may never have, for the present may consume them; they are gender neutral. Rosalind could be any scientist who is seeking perfection. Rosalind is very likely to be a woman.

Leap of faith assumptions, formulating hypotheses, testing them in rapid succession to mitigate risk has become the norm of research – certainly in industry, and, I suspect in academia as well. Much of this approach is driven by the funders and the managers in deference to efficiency and outcome being considered as measures of accountability. An endeavor in which the destination becomes more important than the journey, a scientist who pegs away patiently for the moment of truth is invisible and anachronistic.

To me Rosalind remains an inspiration not only because she was a woman steadfast in her love for nature and science but because she had the desire and the courage to undertake a journey.

 

[i] He and his father William Bragg had been awarded the Nobel in 1915 for proposing and demonstrating the Bragg’s Law of x-ray diffraction by crystalline structures.

[ii] http://www.jagrititheatre.com/photograph-51

Quest for a Fairer World

“In the vastness of space and the immensity of time, it is my joy to share a planet and an epoch with you.”, wrote the astronomer Carl Sagan as a tribute to his wife in “Cosmos”, a book I must have read when I was about 23. Something in his eloquence, the illustrations of multi-hued distant galaxies and my own fascination with the skies had left in my mind an imprint of the universality in his sentiment. It had seemed to me then as it does even today that humility and love for nature and life should be the most natural consequences of the very state of ‘being’.

And yet that is not how it appears. This morning’s newspaper describes a survivor’s account of how a group of men attacked and killed his friend, a man who was bound homewards carrying his cattle – his cow.

I read “Bheda” yesterday. A novella originally written in Odia by Akhila Naik in 2008 was subsequently translated into English by Raj Kumar and has been published by the Oxford University Press. In the Kindle version that I read, there are Notes from the author and the translator that set the context for both the story and their collaboration. A detailed Introduction to the main text is an incisive analysis of the dominant literary discourse and the nature of Odia society over the last few centuries. Bheda is remarkable for the simplicity and the crispness with which it portrays the many aspects of caste oppression. That access to education is necessary for socio-economic mobility is well understood and accepted.  In the story, Dinamastre, an educated Dalit and Dom by caste is respected as a teacher. His family has a reasonably comfortable life. His wife Mastrani is literate and has wider access to the resources in their village. His son Laltu has confidence borne of knowledge and intelligence. But is education sufficient to transcend realities of exclusion? Laltu sees the differential treatment meted out to students based on caste in his school, refuses to accept the status quo and protests. He forgoes the path of scholarship to form the Sahajkol Jungle Suraksha Committee with his friend Kartik and other boys from neighbouring villages to confront the combination of the the rich land owners (Baya) and business men (Semi Seth) who deprive the forests of Kalahandi of timber and workers of their rightful wages. Laltu also openly defies the stated scriptures, deities and blind beliefs that keep the villagers tethered to continued cycles of exploitation. Although it is the men from the upper castes who lead a riot that heinously injures Manu, and sets fire to his modest tailoring shop; all the men of position, power and wealth gang up against Laltu and falsely accuse him of having thrown a bone of a cow inside the local Mahadev temple that catalyzed the riot. At the end of the story Laltu is sent to jail.

Why must an animal be invoked over and over again in this bizarre manner for men to settle score with other men over rights to co-exist as equals in a free country?

And why must those who help and support the oppressed, fight and protest what is unfair and irrational be held to ransom?

Chidanand Rajghatta writes about Gauri Lankesh and her immense grace.  Gauri, whose life he recreated through “Illiberal India” where I came face-to-face with a woman of spirit, courage and conviction who was gunned down.

And the media today resounds with Sudha Bharadwaj, a teacher, lawyer and an activist who has spent over three decades with workers and adivasis in Chattisgarh in a long arduous march towards a more just, safe and equitable future for them. [i]

I did not know Gauri. But, I knew Sudha. How can I ever forget her? She was my senior by a year in school and we both had the same Mathematics teacher. Grace Browne was at once stern and affectionate. She had taught us algebraic equations; she was my best Math teacher ever. Sudha was a senior and a hostel mate at IIT Kanpur too. I don’t remember discussing Grace Browne with her at IIT. But I can still hear her rich resonant voice as she sang “Woh subah kabhi to ayegi …” [ii] in the depth of the night.

 

[i] https://thewire.in/rights/a-woman-must-develop-her-own-identity-and-not-be-subsumed-by-the-collective

[ii] https://www.youtube.com/watch?v=a3g1nivv0-A

 

As I see it

How would have the night looked to our ancestors before they figured out how to rub the flint and produce fire? Imagine an Amazonian forest full of robots who have just been trained by their mentors (humans skilled at mechatronics) to perform the same task that the early homo erectus may have undertaken with a mix of hope, chance and curiosity a million years back. Will the future trajectory of this community of robots alter for having learnt to build fire?

I have been reflecting on the idea and business of artificial intelligence (AI). The buzz is impossible to ignore. In the past week itself there were two different conferences in Bengaluru on the same subject. I had several ex-colleagues and friends giving talks, participating in and moderating panel discussions and amongst the audience along with students, engineers, business leaders, start-up entrepreneurs and venture capitalists.

One way of dividing up inventions is to look at things that humans can do and those that they cannot do. Humans can move but not quite at the pace of car or a train. Humans cannot fly like an aircraft. Humans can see objects of a certain size up to a certain distance. Humans cannot see the structure of a DNA and galaxies that are far away without designing microscopes and telescopes. I can use a pestle and mortar to ground the spices to my preferred granularity but an electric grinder can do it quicker.

Of course we still have to eat and drink for ourselves, we have not yet invented another that can eat for us so that we may live and thrive.

In some ways I am a dinosaur. My favorite notion of AI is an absolutely delightful robot called Bidhushekhar who flew with Newton, the cat and Professor Shonku to Mars in Satyajit Ray’s fascinating tale of space travel, a book that I read in the 1970’s.

In another way, I am contemporary and a gnostic. This is in the context of tasks that we can perform but we let inventions do it for us. Fans, washing machines, bicycles, tractors, calculators; the list is long. They are always faster, more reliable and save us physical effort. It is in this context, the use of computer algorithms, touch screens or mouse clicks to ease tasks that Radiologists perform such as labeling anatomical structures on scanned images and drawing their boundaries in a process called segmentation to measure distances, areas or volumes seem useful as long as the Radiologist has the option to perform the task herself if she choses to or in the event of a failure.

It was in 2011 that my colleagues and I had the chance to test the reliability of such an algorithm which sought to automatically detect the contour of the head of a fetus phantom (object with the structure of a fetus made of material that mimics acoustic properties of biological tissue) from two-dimensional ultrasound images and calculate its circumference. Based on 184 scans, we had found the automated measurements to be consistently more accurate and show lower variability compared to both non-expert and expert operators. [i]

So when an old friend shared the news of a California-based company [ii] to have received FDA clearance for automated measurement of ejection fraction from echocardiography (ultrasound images or videos of the chambers of the heart), I was both pleased and pleasantly surprised. Pleased because it validated our findings regarding automated measurements (that we conducted in a much more constrained experiment) with in-vivo scans; surprised because while staking claims for higher speed or ease of use is not unusual, making automated screening a routine procedure is a disruptive step towards AI-based Radiology, its implications far-reaching for patients, clinicians and regulators.

The procedure implemented by Bay Labs employs Deep Learning (DL) technology and is based on curated data of 4,000,000 images from 9000 patients.

This is where I turn a skeptic. I am not a trained Computer Scientist and I am not what they call a Data Scientist. I am not even sure what that means. However, I have worked in a lab full of experts in medical image analytics who I watched as they transitioned from traditional methods (morphological filters, shape models, active contours, active appearance) to those that were more data intensive (Ada boost, random forests, restrictive Boltzmann machine, support vector machine) and finally the panacea of all problems in detection/recognition/classification aka DL.

I am a Physicist by training and like others of my ilk in industry, used to generating data via designed experiments or simulations and fitted regression models to describe physical phenomena in a ‘finite’ and well-defined design space with the twin objective of designing imaging systems and predicting their performance.

From hearing and seeing what people around me say and practice, it appears to me that DL is a key element to realizing AI.

From what I understand, accuracy of DL-based prediction improves as the training data that is used in building the model approaches a very large number N. For image data from diagnostic scanners (MRI, CT, Ultrasound), N typically represents number of patients. It is not unusual in this field to twist and turn images, dim and darken their parts to generate variants that are most likely unreal but very few appear to care whether they are real or unlikely, very often they do even know the difference between the two. And, why, so?

This brings me to my first concern. In one of my first classes of Elementary Statistics in High School, I was taught to inspect every data before I started analysis. As N becomes large, at best we use algorithms to ‘see’ it; at worst, we don’t see it all, we just feed it into a neural net.

My second concern: in case, we ‘see’ it with algorithms, what do we ‘measure’? How would we represent say an image of a normal liver tissue in an ultrasound image? How do we distinguish it from normal liver of another subject? How do we know that we have captured ‘all’ possible normal liver parenchyma? There appears to be a dearth of metrics to assess representation.

My third concern: we do not care about the imaging device and protocol. The interaction between these and the state of the human subject is complex and nonlinear. We assume that large N somehow takes care of this function.

My fourth concern: we do not care about the operator; a cause of considerable variability in ultrasound images. Once again, we assume that large N somehow takes care of this function.

In building traditional regression models of physical processes, the choice of a full factorial design is fairly standard when the number of variables are few (~3) and the number of their states are few (~3). However, as the numbers of variables or their states grow, we venture into the world of Design of Experiments (DOE). There are a large number of graphical tools and statistical tests recommended to study the behavior of the regression function before a choice is made for the best design. Since DL does finally represent a nonlinear regression model, how do we study its behavior at representative points in the design space to ensure that the model is robust?

For an image of size 100 X 100, each pixel with 256 possible states, a full factorial design would imply N = 25610000. In reality, there are much fewer variants. There are anatomical, pathological, contextual features, background and foreground, patient habitus, consequences of device, protocol and operator (in ultrasound) that need to be taken into consideration in assessing the size of N. I have not seen anyone trying to do this (I am an outsider to the field. May be there are researchers who do this and I am ignorant. If so, I apologise).

In the way DL is often employed, it seems to me that regression is performed without knowing the design space and without applying the fundamentals of DOE. If we could find a way of addressing this, may be it will become possible to better predict the capability and limits of models for enabling AI?

***********

What then would be the future of the robots who can now ignite a fire?

[i] http://spie.org/Publications/Proceedings/Paper/10.1117/12.2008032

[ii] https://www.businesswire.com/news/home/20180619005552/en/Bay-Labs%E2%80%99-EchoMD-AutoEF-Software-Receives-FDA

 

 

 

 

 

The Business of Identity

 

Last week I got yet another business card in my kitty. They make drones, said my friend from across the crowded lunch table as he greeted the gentleman who was seated next to him. I smiled and asked the gentleman what was the name of his company. He nodded briefly and handed me his card.

I have often wondered why business cards continue to be in vogue in today’s intensely digitised and virtual world. What is this curious stickiness about a rectangular piece of material (mostly paper of varying thickness, texture and carbon footprint) with an imprint of one’s identity, affiliation and coordinates?

Is it possible that people carry and distribute the card around because it speaks for them? How we introduce ourselves says so much about us and our sense of self with the nature and milieu of a new interaction. Cards perhaps provide both opportunity for acquaintanceship and respite from yet another conversation?

In my last seventeen years in the corporation, my job title changed four times, the name of my organisation was modified eight times, my office email and phone number altered two times. What remained constant was my name.

But, then something unexpected happened in the midst of all this … 

My first ever business card

Came at the turn of the millennium

I was seven and thirty

And had just joined GE.

I still remember

Those days

When there was no TCP[i]

Project Leader

That was me!

First in Electronics Systems Lab

Then in Virtual Imaging

Within the Imaging Technologies Lab

That we loved and called

The ImT.

Then came the

Biomedical Signal Analysis Lab

In CSSP[ii]

First inside DBT[iii]

Then inside DIBT[iv].

What then was my identity?

TCP tried

Ever so hard

From lead

Through seniority

To establish

My principality.

For the longest time

My title

Was virtually set in stone

In the e-world

It populated itself

When I placed my SSO[v].

Until one day

In twenty-thirteen

I discovered

Quite suddenly

That the title

Had been freed

From the binds

Of the TCP.

With eager rapidity

My fingers

On the keyboard

Could not resist

Typing in

Physicist.

That happened to be

May Day twenty-fourteen

When I was liberated at last

From the need of

A New Business Card.

[i] Technology Career Path

[ii] Clinical Systems and Signal Processing

[iii] Diagnostic and Biomedical Technologies

[iv] Diagnostic Imaging and Biomedical Technologies

[v] Single sign-on

Don’t miss checking out who may have invented business cards:

http://www.designfloat.com/blog/2012/04/02/history-business-cards/

 

With the Rays in the other Bangalore

A rare occasion found me on Cunningham Road in the morning of 9th June. I have now lived in Bengaluru for 22 years. That is longer than I have lived anywhere else. People often ask me “Where is your Native?” I tell them “Home is where I live, here in Bengaluru”.

There is something romantic and lingering about old Bengaluru. Cunningham Road brought back two distinct memories. The first was the forays Rishi and I used to make to the Pizza Hut in our early years – a 35-minute drive from the IIM in the days when Bangalore was mentally bound by the South End Circle. The day I was sketching ultrasound propagation curves on the paper napkin and not paying attention to Rishi may have been the turning point. Somewhere we lost the innocence of a relatively carefree life. My second memory of Cunningham Road is associated with my trips to Mahavir Jain Hospital to meet with Dr. Satish Govind. During our last visit together there Srik, Nav and I drank the second flush Darjeeling tea in a Fancy Tea Boutique that had come up. By then, I was on an aspirational ascent.

This year on June 9 my mother and I went there to watch a Tribute to Satyajit Ray[i] in Chamundeshwari Studios. We reached early and realized that to get some coffee, we would have to walk across Cunningham Road at the traffic signal. The buses and cars rolled in at top speed but came to a complete halt like the ‘civilized’ when the lights turned red in a manner that was sharply different from the vehicles that continuously inch forward at the traffic intersections of ‘new’ Bengaluru where I live. It occurred to me that there are two cities here.

We drank filter coffee at Chai Point where ‘new’ mingled with the ‘old’ at 9.15 a.m.

So, it was in the spirit of the day that Bhakta (the Followers)[ii], Chilekotha (the Attic)[iii], Tollywood-e Tarinikhudo (Tarini Uncle in Tollywood)[iv] [v]and Royal Bengal Rahasya (Mystery of the Royal Bengal) came to be. The first three short films are part of a series of seven that were made for the Television. In fact, they are available for viewing on YouTube. The fourth is a full length feature film. Watching these on a larger screen in a small Studio auditorium reminded me of the late evenings in IIT Kanpur during the mid-80’s when Anu and I would unfailingly show up for the Le Montage films in one of the smaller and cozier Lecture theatres.  I could have been a buff invited to a Preview of a film by a Master that was yet to be released! Besides, I felt drawn to the old-fashioned decor of the auditorium compared to the glitzy PVR Audi’s in Bengaluru Malls.

Satyajit Ray’s stories almost always carry with them a celebration of human endeavor despite human follies and foibles, ambition, greed, unnecessary pride, irrationality, belief in superstitions and a propensity to lie. The four that came alive on the screen with Sandip Ray’s direction – so reminiscent of his father, the music, the movement of camera, the brevity, the quiet humour, a sensitivity beyond the obvious, a certain connectedness with the larger universe, living a life of joy in just being and doing.

This was true of the writer of children’s book who had a large fan following in Bhakta and his look-alike. Sabyasachi Chakraborty, the one who played a double role in this 22:14 minute film came up on stage and gave a brief introduction to the curated film festival and in particular the pleasure he has had in working with Sandip Ray. The man on stage and the two characters in the movie merged into one another seamlessly – metaphysically.

In Chilekotha, the once brilliant Shashanko Sanyal is left lonely and impoverished in Brahmapur village. Poetry remains his only source of emotional sustenance. His once-envious old class mate Adityanath happens to stop by at the village, sees his sorry state, wants to help him and possibly partly atone for his past trespasses. Adityanath … the one who has achieved much more worldly success. The smile on Shanshanko’s prematurely aged face when Adityanath finally rightfully returns to him the silver medal he won at a poetry competition in school is minimalism that moved me to the point of tears.

Tarini Khuro[vi] relates delightful stories to a bunch of children based on experiences from his itinerant and multi-faceted life. Listening to his adventures in Tollywood (Bengali film industry in Kolkata located in Tollygunj is called Tollywood to rhyme with Bollywood, the Bombay Film Industry) along with the kids, it crossed my mind that he could be making it all up! So marvellous was the tale that I would continue to wonder whether life is stranger than fiction or the other way round. His story is both a confession and a matter of great pride to him. For when he is young man and a manager in a film producer’s unit, he impersonates an upcoming actor Ramani Mohan Chatterjee. Tarini itches to play the lead role in a cinematic version of Alamgir. To be able to do that he takes advantage of the superstitious Ramani rendering him incapable of remembering his lines. Ramani lets Tarini act with the promise that the world should not know that it is him. Alamgir becomes an instant hit and Ramani becomes a famous actor. Tarini moves on, happy with the success of Alamgir, generosity in his heart for Ramani Mohan and not an iota of regret.

All these characters, not to forget the wig-maker in Alamgir over whom the camera hovers long enough to catch the loving glance he casts at his creation and Feluda himself in Royal Bengal Rahasya pursue what they deeply care for – they are drawn by creativity, curiosity, drama or an unsolved mystery.

Not fame, name or money.

That day with the Rays I found again the old Bangalore, the old me.

[i] https://www.thehindu.com/entertainment/movies/lost-and-found/article24102656.ece

[ii] https://www.youtube.com/watch?v=jwpDyyP9l9w

[iii] https://www.youtube.com/watch?v=OTmOGyDIkbs&t=98s

[iv] https://www.youtube.com/watch?v=4oKdbgQDd84

[v] https://www.youtube.com/watch?v=Fd7dNYOdoIw

[vi] http://www.liquisearch.com/tarini_khuro/tarini_khuro_series