Often, when doing an experiment for the first time, you don’t know how big the signal is, so you have to make an educated guess at how many repetitions will be required to detect it. The Dog Project was on the verge of moving beyond a cute dog trick and into the realm of legitimate science. But to make this jump, we would first need to figure out how many repetitions would be required.
Andrew and I took a close look at what we had collected on the first scan day. Even though we had failed to find any differences between peas and hot dogs, there was still useful information in the data. We could estimate the SNR of the dog brain and, from that, determine how many repetitions Callie and McKenzie would have to do at the next scan.
Andrew zoomed in on the caudate of McKenzie’s brain. He pulled up a graph of the level of activity in the caudate for each scan that we had acquired. The first few scans had no signal because McKenzie hadn’t placed her head in the head coil until about the twentieth scan. But then it looked like noise. It was hard to tell how much of the noise was because of the usual sources or her moving slightly during the scan. The size of the fluctuations measured about 15 percent of the overall signal. This was much higher than in human studies.
I let out a sigh and said, “We would need a thousand repetitions to get the SNR up to a reasonable level.” Neither dog nor human would sit still for that long. “That has to be from movement.”
“It is,” Andrew said. “Check this out.” Andrew scrolled through the sequence of McKenzie’s images. This had the effect of creating a movie of her brain. It compressed the five-minute scan session into thirty seconds. Even though we had captured only half of her brain, the movie made clear that although McKenzie had been in the head coil for the whole session, she was still moving. Not much. But just enough to cause artifacts. Callie’s brain movie looked similar.
In fairness to the dogs, they had done what we asked of them. The degree of movement we were talking about was a matter of millimeters. During the stress of the scan session, neither Melissa nor I noticed it. Not that we could have done much about it on the fly.
“Well,” I said, “we have two weeks to train them not to move.”
Andrew looked skeptical.
The dogs would have to move less than two millimeters, but they would have to hold still only during the period from putting their head on the chin rest through the duration of the hand signal. After they got their hot dog, they could take their time swallowing and getting resettled in the chin rest. We needed enough time only for the fMRI signal to reach a peak and begin to decay—roughly ten to fifteen seconds. If the dogs remained motionless for that length of time, we calculated that twenty repetitions might be enough to get the SNR up to a usable level. That still left the structural scan, which we hadn’t been able to obtain on either Callie or McKenzie. That scan would require the dogs to stay motionless for thirty seconds.
Thirty became the magic number. During training, we would have to gradually lengthen the time between the hand signal and the reward until the dogs could hold absolutely still for half a minute. If they could do that, we would be able to get the structural scan and plenty of functional repetitions to boot.
Callie didn’t seem to mind the change in training procedures. At first, I felt a twinge of guilt every time I put up the signal for “no hot dog.” Callie would stare at me impassively from her position in the mock head coil. I’m in the head coil, why no hot dog? Sometimes I would touch her lightly on the top of her head, indicating that I wanted her to try again. But this soon became superfluous.
It seemed cruel to withhold rewards, but I trusted Mark’s advice and stuck to the VR2 training schedule.
Mark was right. After switching to variable reinforcement, Callie really started paying attention. She had no choice. With peas and hot dogs, she got rewarded every repetition, so there was no need to pay attention to what I was doing. Now she noticed every little movement. If my shoulder twitched, Callie’s eyes darted to the side. It was so fast that had we not been staring directly at each other I would never have noticed.
I continued recording our training sessions. With a digital camera on a tripod, I could shoot directly over my left shoulder. Even though Callie and I were staring directly at each other during training, the camera picked up things that I hadn’t noticed in real time. Mark and I reviewed these videos like football coaches on the Monday after game day. He critiqued my technique as we tried to eliminate all of my “tells.” We wanted the dogs to be focused solely on the hand signals. Callie wasn’t the only one who would have to hold perfectly still. So would I. Except for the hand signals, we didn’t want any extraneous body movements.
We amped up the noise training too. Both Callie and McKenzie had reacted negatively to the sudden onset of the shimming and localizer sequences, so we incorporated recordings of those noises into the daily training as well. The more the dogs became accustomed to the sounds, the more comfortable they would be.
We even tried to make positive associations with the noise. I would start playing the scanner noise through the PA and call Callie to the living room. We would wrestle and play tug-of-war while the noise blasted away. Lyra would join in too. It took only a few days before Callie would run to the living room as soon as she heard the scanner noise playing. I would slip on the earmuffs and crank the amp to 95 decibels to give the full effect. She didn’t care. Callie would just trot up the steps into the tube and plop down in the head coil, licking her lips and waiting for hot dogs.
It was during this intense training period that I think our relationship began to change. Rather than master-dog, or dominant-subordinate, we became a team. We were like pitcher and catcher. For lack of a better word, it was intimate.
There is something deeply personal about staring directly into another’s eyes. Humans’ eyes are unique. We have more white in our eyes than any other animal, which means that we can tell with extraordinary precision where other people are looking. One theory says that humans’ eyes evolved this way as a means of nonverbal communication. Using nothing but eye movements, we can, for example, communicate to other people where they should direct their attention. Just as important, we can deduce a great deal about someone’s thought processes just by observing where they are looking. Gazing directly at you? They are definitely interested. Gaze averted or roaming? Not so much.
Under normal circumstances, when I had looked into the eyes of animals, even our beloved pets, I never felt a strong reciprocal connection. Sure, they looked back, but the gulf between species was too great. It was like staring into an abyss with no clue as to what lurked behind those big brown eyes.
Now, eyeball-to-eyeball, I could see my reflection in Callie’s eyes. Yes, she wanted hot dogs, but there was something more. Callie had been communicating with me the whole time. I had been the one who was blind to it. But now that we were staring at each other for minutes on end, there was no ignoring it. Subtleties of expression—how she held her eyebrows, the tension in her ears, the drape of her lips, and, of course, where she directed her eyes—spoke volumes.
Now too late, I realized that Newton had done the same.
As dog trainers have known for a century, dogs are exquisitely sensitive to picking up cues in their environment. Dogs act with a theory of behavior, which is the broad scientific term for saying that dogs learn that certain behaviors lead to certain outcomes. This is the foundation of positive reinforcement. But staring into Callie’s eyes, and watching how she stared back, I began to suspect that she was doing something more. She was noticing where my attention went.
The ability of dogs to track others’ attention has only recently been appreciated. In 2004, researchers in Hungary tested the extent to which dogs used attentional cues from humans. They set up a series of experiments that included different types of fetching tasks that varied the face and body positions of the humans. The researchers wanted to know how dogs reacted to a human when they either faced each other or faced away and whether the visibility of the human
’s eyes made a difference. To hide the human’s eyes from the dogs, the person was blindfolded. The researchers found that dogs were sensitive to the human’s attention, but that it depended on the specific context. In tasks that were playlike, the dogs didn’t seem to care whether the human was looking at them, but if the human commanded the task, then the dogs paid close attention to where the human was looking.
The evidence continues to accumulate that not only are dogs sensitive to where humans’ attention is directed, but dogs are also sensitive to the social context. They know when it is appropriate to attend to their human’s attention and when it is not. This means that dogs have more than a theory of behavior. They have a theory of mind.
In humans, theory of mind, or ToM, means that we can imagine what another person might be thinking. Reflecting the importance of humans’ social lives, most of our large frontal lobes seem to be concerned with this function. We spend huge amounts of mental energy navigating the complex social structure of human society. Knowing how to read people and how to behave in distinct social settings is the difference between success and failure. And at the extreme, autism may represent a failure of the ToM system in the brain.
If dogs have ToM abilities, they are probably simpler than ours. The small frontal lobes in the dogs’ brains are clear evidence of that. But even if dogs have only a rudimentary ToM, that would mean dogs are not just Pavlovian stimulus-response machines. It would mean that dogs might have about the same level of consciousness as a young child.
As Callie and I honed our performance, I had a growing sense that we were beginning to read each other’s mind. Of course, there was no way to prove this. The thought was so crazy I didn’t even voice it in the lab. But we were about to discover that my intuition wasn’t completely off the mark.
The second scan day arrived on a drizzly February afternoon. Under cover of umbrellas, the entourage once again made the trek from lab to hospital. The novelty had faded somewhat, so fewer people were in attendance, and the overall atmosphere was calmer and more businesslike. Robert and Sinyeob greeted us at the scanner. This time they weren’t laughing. Everyone knew the dogs could do this, and we were there to do science.
There was no need to fiddle with the scanner settings. Robert simply pulled up the final parameters from last time, and we were good to go. The plan was to do the shimming and localizer, two five-minute functional runs of hot dogs versus no hot dogs, and then a thirty-second structural scan. If there were no hiccups, we could blaze through the procedure in thirty minutes for each dog.
It really helped that everyone knew what to do now. Rebeccah worked her magic touch with the earmuffs and head wrap on Callie. Andrew took up his position at the rear of the magnet, ready to record the repetition type—hot dog or no hot dog. Melissa and Mark settled McKenzie in her pup tent until it was her turn. I motioned to Callie to go into the scanner.
To avoid startling the dogs with the sudden onset of buzzing, Mark had hit on the great idea of playing the recordings that we had used during the training procedure. Every MRI has an intercom to allow communication between the patient and the technician. After Callie got settled in the chin rest, the team in the control room held an MP3 player up to the intercom and began playing the recording of the localizer noise. Softly at first, then they gradually cranked up the volume. Pretty soon I could hear the familiar swarm of bees emanating from the speakers built into the magnet. Because it came on gradually, Callie didn’t budge.
Callie’s localizer with box indicating field of view.
(Gregory Berns)
I nodded to Andrew. The buzzing continued. And then it stopped.
“What happened?” I asked. Andrew shrugged. Callie followed me into the control room. “Why did you stop the scan?”
Robert looked confused. “We didn’t,” he said. “Look.”
There, on the computer screen, was a perfect image of Callie in profile. The image was sliced right down the middle of her head, giving a beautiful view of her brain and spinal cord. Robert had already placed the bounding box for the field of view in place. The use of the recording through the intercom had worked so well that neither Callie nor I had noticed when the real scan started!
With the field of view set, we cued up the functional runs. I showed Callie the container of hot dogs, and her eyes widened. All I had to do was point to the magnet, and she scooted in.
This time, we played the recordings from the functional sequence through the intercom. The volume was gradually increased, and after a few seconds, I could hear the real scans begin. They sounded almost identical. Callie didn’t care. Her eyes were lasers on mine. I held up my left hand to indicate that she had done well and gave her a piece of hot dog.
We were off and running. I alternated repetitions for hot dogs and no hot dogs but kept it somewhat unpredictable, throwing in runs of two or three of the same trial type. Callie stayed cool as a cucumber. Every time I put up the sign for “no hot dog,” she stared at me and waited until I put up the sign for “hot dog.” I began to appreciate that rather than being disappointed during the no hot dog trials, Callie viewed those hand signals as uninformative. Being told that she wouldn’t get a hot dog said nothing about when she would get one. This interpretation would soon be borne out by her brain activation.
Unlike the previous scan session, this time we were much more efficient. In short order, we had acquired two five-minute runs of functional scans, nearly four hundred images in total. The only thing that remained was the thirty-second structural scan. At this point, Callie looked either tired or bored, but in she went for the fourth time. The recordings of the structural sequence were slowly ramped up through the intercom, and then the real scan started. The structural sequence sounds much like the localizer, but Callie stayed put throughout.
She had done it. She hadn’t moved at all. I ran around the scanner and gave her a handful of hot dog.
“You are such a good girl!” I exclaimed. “You are a SuperFeist!”
Robert already had the structurals on the screen. There, in breathtaking clarity, was the first detailed structural image of a completely awake dog. My jaw dropped. We had just acquired nearly four hundred functional scans and a structural image that rivaled anything we got in humans.
The first detailed structural image of Callie’s brain rivals the quality of human scans.
(Gregory Berns)
Even if McKenzie bombed on her turn, I was confident that we had achieved our goal of getting enough functional scans.
“How many repetitions did we get?”
“It looks like she did twenty trials with hot dog and nineteen trials of no hot dog,” Andrew said.
“Damn,” I marveled. “That should be plenty for analysis. Let’s hope the SNR is high enough.”
Callie sat down next to me. I looked into her eyes, and she knew. Yeah, I’m the top dog.
If I’d had any lingering concerns about Melissa and McKenzie, those quickly disappeared. The trick of playing the recordings through the intercom worked wonders for them too. We finally got a localizer image for McKenzie, which allowed us to precisely place the field of view to avoid chopping off half her brain this time. For the functional scans, Melissa was more collected than I had been. She really took her time with the repetitions, requiring McKenzie to hold still for fifteen seconds for each trial, where I had required Callie to hold still for only ten.
McKenzie was like a rock. Robert and I watched her images stream on the console in real time. She was not moving. Not at all. They blazed through the two functional runs, and, for the first time, we got a structural image of McKenzie’s brain.
Two for two.
Not only was the day a complete success, but we had accomplished all of this in two hours—half the time of the previous session.
It was still exhausting. When you’re locked face-to-face inside the magnet with jackhammers all around you, the level of concentration, for both dogs and humans, is intense. When Callie and I finally got home
, we crashed together on the couch. We looked at each other once and then closed our eyes.
19
Eureka!
ANDREW DIDN’T WASTE ANY TIME. The next day, he had already begun the analysis of Callie’s and McKenzie’s data. Just like the peas and hot dogs experiment, the first and trickiest part of the analysis would be the motion correction. We had to carefully identify which scans contained brains and discard the ones in which the dogs moved too much. Animating the sequence of images in rapid speed helped make the task easier.
Andrew showed me the animation.
“Check this out,” he said. A pixelated image of a dog’s brain danced on the computer screen. For stretches of several frames, which were actually tens of seconds in real time, the image didn’t move. Except the eyeballs, which darted left and right.
“This is Callie,” Andrew continued. “She did really well. If we throw out the scans with movement artifacts, we still have 62 percent left for analysis.” My heart swelled in pride at my beloved feist.
“That’s amazing,” I said. “That is five times better than the previous session. How about McKenzie?”
“Almost as good. We can keep 58 percent. She had sixteen hot dog trials and eleven no hot dog trials.”
“Melissa was really making her hold still for a long time,” I said.
“Yes,” Andrew said, “but that means we’ll have a lot of scans for each repetition.”
We spent the next two days checking and rechecking each step of the analysis. To make sure that we didn’t mistakenly confuse brain activation with movement artifacts, we kept ratcheting up our criteria for whether to keep a scan in the analysis. Andrew and I would stare at the animations, looking for even the slightest twitch of the head. Most of the head motion occurred when we gave the dogs hot dogs. This was no surprise. But we weren’t interested in the brain response to hot dogs. We were interested in the response to the hand signals. When we were satisfied that we had identified and discarded all the scans with motion, the remaining scans showed that the dogs had held their heads with less than one millimeter of movement during the critical period of the hand signals. That was as good as humans do in the scanner. We were ready for the final step: comparing the activation between the two hand signals.
How Dogs Love Us: A Neuroscientist and His Adopted Dog Decode the Canine Brain Page 16