We Induced Smells With Ultrasound

We pointed an ultrasound probe at an odor-processing area of ​​the brain to elicit different sensations. Different focal spots correspond to different odors, which we repeated on two people in a first attempt and validated with a blind test. The sensations we received are:

Here’s a video of our blind testing:

Net

Odors are processed in the olfactory bulb. We decided to try stimulating it with focused ultrasound through the skull. As far as we know, It seems that no one has done this type of olfactory stimulation before – Even in animals.
However, after being able to induce sensations of motion last week, attempting the same for olfaction seemed promising.

Anatomy

The olfactory bulb, our target, is hidden behind the top of the nose. This has become quite an inconvenient place for a few reasons:

1. The nose does not provide a flat surface to place the transducer for stimulation.
2. It is mostly filled with air, which interferes with ultrasound. Ultrasound requires a continuous medium to propagate, and filling the nose with gel seemed distasteful.

Instead, we found that you can place the transducer on the forehead and aim the ultrasound downwards toward the olfactory bulb. Although this is not a perfect solution as the frontal sinuses can weaken the signal, careful positioning of the device over the sinuses still allows us to reach our general target area.

ultrasound

We got our first impressions using just a handheld probe and some gel, but it quickly became clear that holding the probe steady by hand made it nearly impossible to keep the focal spot in the same location. To improve stability, we improved a floating headset, allowing more reliable positioning. We switched from gel to solid, jello-like pads for stability and general comfort. Finally, our headset got a little hack:

Eventually a knife was stuck into the probe for mechanical assistance. At the time, all of our headsets had a knife stuck in them for testing, because opening the knife would cause software errors.At some point we thought about using a mouthguard to fix the probe relative to the brain, This was a good idea because the teeth are the only exposed part of the skull, except that you can’t talk about smell while wearing a mouthguard,

To guide placement, we used MRI of Lev’s skull to determine roughly where the transducer would point and how the focal area (where the ultrasound waves actually focus) aligned with the olfactory bulb (the target for stimulation).

We found that our “sweet spot” is low-frequency ultrasound focused just below the forehead and angled downwards toward the bulbs. especially:

• 300 KHz frequency (low enough to penetrate the scalp well)
• Focal depth of approximately 39 mm (where ultrasound energy converges below the forehead)
• 50-55° steering angle (to point the focus down towards the bulbs)
• 5-cycle pulses (short, rapidly repeating bursts) at a 1200 Hz repetition rate

While Albert did not have an MRI available, this general configuration with minor adjustments to focal spot position still worked for him.

Security

The largest portion of time was spent ensuring that the ultrasound sequences split in two directions safely and as we expected:

1. Measuring the output field. We placed the transducer in the water tank and measured the pressure at the focal location. With our parameters, it ranged from 150 to 250 kPa, which corresponded to a maximum mechanical index of 0.4. This means that the average intensity at the focal spot was lower than that typically used in TFUS and has been proven to be safe. We were also within safety limits on mechanical index and thermal dose.
2. Avoiding the optic nerve by reducing asymmetry in the system: The nerves lie beyond the hairline. The two components of the olfactory bulb are also slightly off center, so a slight asymmetry was necessary: ​​we focused at a 2 degree angle to the edge in one of the presets. However, we stayed within the 15 degree range, which is enough to not touch the optic nerves.

Result

We have managed to generate four different sensations, all of them in two people:

we distinguish between a smell And a feeling Here because, subjectively, they feel different. The smell is strong and localized in line with the noise, almost in the same way you can sniff around and locate the source. Sensations are more diffuse: a weak, slow-onset impression of smell, often paired with other (potentially placebo) feelings, such as a slight tingling sensation on the face.

Both smell and sensations are most intense when lightly inhaled, so we conducted the test by sitting there, placing the probe on the forehead, and sniffing lightly. Sometimes a slight waft of scent takes just a few breaths, and sometimes it just hits you. When Albert first smelled the garbage, he opened his eyes with a jerk and thought a garbage truck had just arrived! It was inside the house.

Many of these odorants correspond not to specific receptor types but to combinations of receptors. We think this is because the focal spot is quite large – the wavelength of 300kHz ultrasound in tissue is 5 mm, whereas the length of the adult human olfactory bulb is about 6–14 mm. Olfactory bulb size can vary up to 3x depending on “age and olfactory experience”, so perhaps (we’re making this up) with more use your olfactory bulb may actually become larger, leading to better resolution stimulation!,

We detected different odors by rotating the beam ~14 mm (20 degrees at a 4 cm radius). The distance between freshness and burning was ~3.5 mm. We ensured that the effect was not placebo with an auditory mask (playing music through AirPods), so you do not hear the probe, although you cannot distinguish different focal spots through sound. We then tested discrimination in a test where Thomas selected focal locations, and Lev was naming scents. You can watch the complete video here.

It is remarkable that we can induce different odors with such small steering (40% of the diffraction-limited focal spot size) And possibly even more, because there was some empty space in between the focal spots, where you couldn’t feel anything.This shows that the resolution we have access to is much greater than the spatial resolution of (kind of) ultrasound Super-resolution for Neurostim!) In particular, we do not need single-neuron resolution to find an independent base of odors on which we can build our latent space. To improve this system, the next steps are a more stable setup, increased frequency, focal location, spot size, and more playing with the excitation waveform.

can you feel the meaning

The interesting reason for stimulating olfactory sensations is not just “VR for smell”, as one might initially assume. have a nose 400 distinct receptor typesAnd we can isolate the subtle modulations of their activation, so they can serve as a channel to write directly into the brain, as a means of non-invasive neuromodulation.

The olfactory system potentially allows for 400 if not 800 sounds due to the two nostrils in the brain. This is equivalent to the dimensionality of LLM’s latent spaces, meaning you can appropriately encode the meaning of a paragraph into a 400-dimensional vector. If you have a device that allows this type of writing, you can learn to associate input patterns with their corresponding meanings. After that, you can directly sniff the latent spaceA little ultrasound, a little breathing – and you’ve got a paragraph,

People are able to develop synesthesia – being able to hear colors and see smells, and it may be possible to extend this to semantics. However, this is speculation at this stage.

One could try to make a similar argument for the eye: take 400 cones on the retina, hijack them, and you get a 400-dimensional channel. But we think the nose is better. olfactory system is excess Simpler and more direct interface with key brain areas such as the hippocampus. The signal is simply less filtered and processed through the olfactory system. If you tried to write arbitrary light intensities into a patch of cones, the next step of processing would be a convolutional neural network-like structure in the visual cortex, and the signal would be averaged. The embeddings you write will never reach higher levels of processing in the brain. You could try encoding the information in a more easily understandable way, such as Chernoff faces, but this will reduce bandwidth, and learning the remapping will still be very difficult.

In contrast, only a few synapses separate olfactory receptors from the hippocampus. This is why certain smells bring up such strong memories!which is responsible for memory, as well as the amygdala, which carries out emotional regulation.

Finally, personally speaking, writers use their eyes and ears more than their nose during office work Raphael Hotter said this is actually a general statement, as the use of eyes and ears goes beyond office work.The nose is a less used channel that implements less poor priors (spatial/tonal maps) than visual, auditory and somatosensory,

We got four fragrances in a few days. With a little more engineering, it should be possible to increase the bit rate of olfactory stimulation Very,

If we gain control of all 400 basis vectors, we may be able to meaning of smell,
And we’ve already covered the first one percent.

approvals

We thank Raffi Hotter, Aidan Smith, and especially Mason Wang for thoughtful feedback on this blogpost.