An Erratic History and A Gleaming Future
Figure 1 Courtesy of the Boston Globe
Figure 2 A bistable image of an industrial sponge with cutout holes, 1970. Courtesy of J. Birnholz, MD
Figure 3 10mm pseudo-nodule with 5mm nucleus in a commercial elastography phantom (Sonoscanner U-Lite, 38mm linear array, 10MHz center freq.) Courtesy of Jason Birnholz, MD
Gray Scale has been the universally proclaimed description of ultrasound imaging for 40+ of my 50 years’ involvement with this field. It is a minefield of paradoxes. There is no facet of ultrasound deemed so basic, yet is so unused; acknowledged, yet misunderstood; necessary, yet unreliable.
Plane X-ray and photo images are overwhelming barrages of vast armies of regimented, infinitesimally tiny, photonic bullets traveling in concert at light speed. There are few interactions between X-Ray photons and tissues, but they are clean and predictable. Tissue information is limited, but reliable and repeatable.
The ultrasound situation is very different. There are rag tag, marsh mellow wisps of phonons that are big, slow, unruly, dim, and easily distracted as they creep along the tight lattice works of body tissues. There are multiple, messy scattering events that generate a lot of noise but which also convey information specific to the macrostructure of what the beam illuminates and that we see as a gray scale image. In terms of information, plane x-rays or still photographs involve a gigantic number of photons, while a phonon impoverished ultrasound image is barely a grain of sand on the same statistical beach.
“Seeing” a lesion or bit of anatomy depends on its inherent “contrast,” which is compromised by noise. The gray scale conveys the contrast features of the ultrasound field.
A Hairy Portable Exam
I Started to think about gray scale when I found a reminder of a visit to the Stoneham Zoo around 1980 to see Gigi the gorilla, who was so excited by the prospect of her first ultrasound that she needed to be sedated. That was convenient for scanner peace of mind too.
This study had to be done on-site. Portable and mobile forms of ultrasound equipment have been around since before 1975, but they have never been popular. Hospitals tend to centralize ultrasound facilities and delegate some of their primary equipment for remote studies in intensive care locations. Mobile ultrasound services have never received much academic recognition, nor have they tended to contribute to ultrasound literature. The thoughts that came together were old and new point of care ultrasound equipment and gray scale capabilities then and now, and the main sources of information being sought in ultrasound images.
Fetal Femur Measurements
Penetration and multiple reverberation noise was a big concern for doing Gigi. Fat is bad, a lot of fat and a lot of muscle are worse. Obstetric ultrasound was one of the original main application areas, because the target structures were large bones that are easy to see, which is to say they are high contrast targets Lack of ionizing radiation exposure was a factor too, but doesn’t get to first base, if you cannot resolve the target against the background. As you would expect, Gigi had a smaller cranial volume and disproportionately longer arms and forearms than a human fetus at the same stage of pregnancy.
The most accurate way to measure the shaft length of the femur is with the bone parallel to the face of a linear array, because measurements made this way do not involve any assumption about sound velocity. When you make measurements properly, you find that there is not only a very tight correlation between femoral shaft length and gestational age, but there is essentially no variation until pretty late in the third trimester. This is a straightforward task that done about a zillion times a day around the world with almost any kind of equipment. GA should be nailed down to within a day or two from this one datum, but that is still not common practice.
The Zebra Paradigm
In retrospect, I think the combined meeting of the AIUM and the WFUMB in San Francisco in 1976 was the best big ultrasound meeting I ever attended. I was living and working nearby at Stanford, and I got in for free by presenting five scientific papers. One of these was entitled The Zebra Paradigm, which evaluated white on black vs black on white TV display options. Some of the influential early ultrasounders used a black on white display for their newly acquired gray scale replacements, ignoring well established principles of perceptual psychology and physiology. Display phase for ultrasound is just like seeing stars at night, but not during the day. For an effective gray scale display, you must have a white on black display, and that has to go from jet black to bright white.
The emergence of GRAY scale Ultrasound
The papers I gave at that meeting were all from new developments the previous year, when I had my first real post-residency job. I was in the right place at the right time, and I had enough technical education and clinical exposure to appreciate the advances sprinkling out of Silicon Valley. I had also just been let out of MIT, where I enrolled so I could learn how to extract tissue specific information from ultrasound signals.
The Searle Pho/Sonic was a manual scan device, i.e., the transducer was affixed to an articulated arm with a position sensor, constrained to a line path determined by the operator. The path and field of view could be across or along the abdomen or just the width of a nodule within an organ. The articulated arm was a marvel. It could be positioned and angulated easily, and it might have been big enough for a gorilla to hang onto without bending. Probe wobble and position sensing were impediments to scanning with earlier units. There was a pre-amplifier within the transducer housing, so low level signals were not lost. The transducer had a large circular aperture, was mechanically damped, and had matching layers, which resulted in a reasonably broad band signal and a lengthy focal zone that was narrow around the beam path.
The biggest improvement was the display, recorded echo amplitudes within an image frame to be read by TV camera and displayed as shades of gray. There were tools like an histogram of gray scale levels within a region of interest, and there were multiple options for gray scale assignment.
The images were spectacular. There was a visual richness to tissue features and parenchymal fields that I had never seen before. Because reflectivity is related to the distribution and type of collagen within the scanning field, histopathologic correlations were direct and in reverse Pathology textbooks could be used to identify lesions that were outside your experience.
The down side was that scanning took a long time. First, you made a test swipe and then reset the controls and kept on hunting and pecking and taking images until you decided you were done. Prior to the Pho/Sonic, it took long enough to scan, and the information was pretty minimal, so, many institutions used technologists and had them follow a protocol of making a series of longitudinal and transverse scans for someone to look at later. This common scheme shifts the nature of an ultrasound exam from making a diagnosis to making a series of pictures.
High Speed Ultrasound
Another big local event of 1975 was the commercial release of the Varian Electronic Sector Scanner, i.e. the phased array. The beam was steered electronically by adjusting the timing of excitation of each element for every pulse launch. Scanning speed was fast enough for flicker free viewing, the display format was pie shaped. Initially, there was a 3 MHz transducer, designed for intercostal viewing of the heart. The unit was self-contained without the need for any position sensing arm. It was on wheels, and weighed about 400 pounds: mobile, but not at all portable.
I became fond of the Pho/Sonic with use, but the Varian was love at first sight. I could move the probe around wherever I wanted, and, because the imaging was continuous, I could adjust the unit for best image quality while I was looking at the screen. To me, image quality has to do with finding or excluding some kind of pathology, so it can vary within and between exams. When probe position changes, geometry changes, adjustment tweaks are necessary. The small probe size was great for catching the liver, spleen kidneys and adrenals in a wink. A limited field of view was balanced by moving the probe freely.
The gray scale display of the Varian was OK, but not fabulous. So, what I did then and afterwards at Harvard, was to adapt standard X-ray fluoroscopy practice. I and a resident or fellow inspected every patient with the Varian and got spot films. Then, that operator or a technologist used the Pho/Sonic to make detailed, overhead pix of selected features. Each exam was a diagnostic consultation. The examiner checked out whatever he or she thought was in the patient’s best interest. This is the ultrasound mirror image of limiting X-Ray or isotope studies to prevent unnecessary ionizing radiation exposure.
Both units were revolutionary, neither had much popular success. There were not many centers with either unit, and I doubt there was any other place that routinely used both for the majority of exams. The next major advance was the release of the Acuson 128, which I tested in 1982 and which was upgraded and eventually replaced by the phase correcting Sequoia. Acuson introduced dynamic control of beam forming on send and receive. One of its special features was to repurpose the array for small areas of interest, which is exactly like surveying by eyeball then slipping on a special lens for a higher resolution microscopic view. The reason the Acuson was great and the Sequoia, greater was in noise reduction. There are many kinds of noise; I am using the term to include everything which degrades information in the imaging process.
Cysts vs Solids
The aim of gray scale is to find the energy range where tissue specific information exists and purify it from noise of all sorts. The original A-Mode was a tracing of echo-amplitude with depth, which is a kind of sonic needle biopsy. The clinical uses were midline echoencephalography and for measuring fetal biparietal diameter at term (vertex presenting). Early B-Mode images used probe position sensing and long persistence CRT displays that only showed the strongest A-Mode signal peaks. This was the ‘bi-stable’ B Mode.
I did image 2 of an industrial sponge with an Aloka manual scanner, which used a signal feedback loop to bring up slightly lower level reflectors. There is no gray scale, because the display scope was not capable. Athersclerotic aortic aneuryms were prevalent then and a diagnostic hot topic. The Aloka identified aneurysms and revealed layers of organized clot, which other units of the era could not do. There is no gray scale to the image, but its information content seems higher than an image limited to just target margins.
Ultrasound people have ever been obsessed with cyst vs solid distinctions. Figure 2 would be a great example if you had industrial sponges for patients. The same crew that were into black and white for early gray scale displays did something far more insidious and unintentionally malign than making a bad 50-50 bet on display phase. They applied the cyst solid concept in extremis, assuming that an organ is solid, so it should be filled with echoes. They promoted a standard practice of doing a lot of signal compression and using a post scan display curve that brought up low level echoes. This seemed plausible esthetically, but it’s a way of filling in gaps with pure noise. If you did this for the liver, you would kiss goodbye to finding primary liver cancers and most metastases, fatty infiltration, or fibrosis. You judge photographs through their emotional content. That’s the last thing you want to do with medical images in general and ultrasound specifically.
Upping the noise replaces the gray scale with pixel density, such as sparse or full fields or ones that are homogeneous or heterogeneous. When you looked at Figure 2, did you think there was something different about the right and left sides of the sponge interior? With the systems of those times, increasing probe frequency also increased noise, leading to strange notions about lesion ‘echogenicity’. Equipment adjustments should be made during scanning, not afterwards on frozen images.
Contrast, contrast, contrast
Contrast is like location to real estate people. If you cannot see a target, it does not matter what your spatial resolution or scanning speed might be. Targets are visible when their contrast with the background or with each other goes above a detection threshold, reiterating intimate relationships between target propertie, sensor performance, and equipment adjustment. An explicit definition is even more elusive, because there are separate contrast concerns, meanings, and factors for every part of the imaging chain, including the eye and brain. Ideally, we want to map tissue features, in reality we have a soup of meat and machine to cope with.
To be fair, the limited gray scale mode of operation is a path of least resistance for the ultrasound community, minimizing technical and educational demands and interpretive challenges by throwing away information that is hard to obtain and seemingly impossible to standardize. The lower the noise, the better image fidelity becomes.
Ultrasound elastography was envisioned by its inventors as a way to up the contrast for a region of tissue at the expense of spatial resolution. This can be great if you have noisy imaging and make the assumption that a lot of things are just not visible. But it unravels for ultrasound, because noise reduction improves both contrast and spatial resolution. The focal length of a lens is effectively doubled for a digital SLR camera versus a film one. Noise reduction, like that possible in broadband digital equipment is like doubling center frequency without losing penetration. A familiar real life example was the introduction of endovaginal probes, which were initially were low frequency. Taking away the multiple-reverberant noise from layers of skin and fat and fascia, made EV pictures much better than transabdominal pix from the same unit at comparable exam frequencies.
Back to the Future
Solid organs really aren’t solid with regard to elastic features, and with every bit of noise reduction, true-to-life images look coarser and more granular than earlier versions. The beautiful thing is that with effective noise reduction, image gray scale features become more robust indicators of tissue substructure and state. The best noise reduction is possible and achieved with dynamic digital control of every step in the imaging process with multi-element piezoelectric broad band transducers. Think of taking a joy ride. Cars have computers that are constantly adjusting things like fuel flow, aeration, and ignition while you appreciate the scenery or smog. The adjustments may be for fuel economy or for performance as identified by the manufacturer. For ultrasound, beneath the hood optimizations are for maximizing diagnostic information and reliability
The development of ultrasound imaging equipment is a story of progressive digital ethnicity. The hearts of new high performance systems are dynamic digital beam formers that modify the system in real time based on the signals that are detected. It is almost biological and it requires a lot of very fast processing power.
Ever hear the term: the three body problem? It is a term for the fiendish difficulty of figure out how three moving bodies interact when you have a lot of information about them and the impossibility when information is sparse. Digital ultrasound devices have control of the entire imaging chain. The problem for designers is that this is a 256+ body problem that has to be solved for healthy and normal tissues and for a range of pathologic deviations. The solutions need to be stabile despite wear and tear, ambient temperature, power, and other external factors. These are complicated devices that deserve to be identified as high-end.
Controls available to the operator of a really smart system can be reduced gain (brightness) and contrast as they relates to acoustic sampling, adjusted during scanning to compensate for differences in patients, like gender, age, hydration, fat and muscle mass.
There can also be adaptive display algorithms that optimize gray scale assignment for human perception frame by frame as the image is reconstructed. This is a way of partitioning echo information across the full visual gray scale range in compartments appropriate to the sensitivity of the eye and brain. This kind of self-normalization makes for consistency in image quality and lends itself to fixing the problems of ageing eyes and yellowing lenses, which I have come to know all too well. The concept of perceptual optimization goes back to Werner Frei: Image enhancement by histogram hyperbolization, Computer Graphics and Image Processing (1977) 6:286-294, way back at the start of voyage up the sonic gray scale river.
High performance digital devices come in different sizes. My own unit is an ultraportable tablet. Figure 3 is one I did on-site at the last RSNA. It is a simulated lesion in a commercial breast elastography phantom that is not supposed to be visible with conventional ultrasound. I am not that crazy about phantoms, they are no substitute ever for patient scanning, but they are cheap to transport and never need to be fed.
What impresses me more than the nodule is the appearance of the material surrounding it. The lack of uniformity is real, identical in repeat views and at alternate slice angles. With a real breast mass, the architecture within the mass tells you something about its identity, while changes at the edges, such as edema or a desmoplastic response indicate the cancer’s invasiveness and the body’s response.
I started with portable equipment, because another facet of digital devices is that low performance devices can be shrunken down to amazingly small sizes. Most of these are low performance devices by current standards, and there is a prejudice that small devices are only capable of minimal, basic tasks, however, there are very few high performance ultraportables now and more may be anticipated. Size affects where exams are done, but performance governs what you do with ultrasound and how well you do it. Paradoxically, the most difficult diagnostic tasks for imaging revolve around early diagnosis, minimal disease, and preventive health, which is the exact opposite of people with more advanced stage disease who require hospitalization.
Medicine is conservative, because safety comes from sticking with what is established and works. Ultrasound is a powerful diagnostic resource that has provided actionable information throughout its history. The problem is that a lot of policies and practices were solidified when equipment performance was less, general medical sophistication and knowledge were less, and societal needs were different. A lot of the world is playing sonic, sandlot baseball. It’s time to move into digital stadia, where smart ultrasound adjusts the playing field for the application game we are playing.
The Reading Room Podcast: Emerging Trends in the Radiology Workforce
February 11th 2022Richard Duszak, MD, and Mina Makary, MD, discuss a number of issues, ranging from demographic trends and NPRPs to physician burnout and medical student recruitment, that figure to impact the radiology workforce now and in the near future.
New Study Examines Short-Term Consistency of Large Language Models in Radiology
November 22nd 2024While GPT-4 demonstrated higher overall accuracy than other large language models in answering ACR Diagnostic in Training Exam multiple-choice questions, researchers noted an eight percent decrease in GPT-4’s accuracy rate from the first month to the third month of the study.
New Study Examines Agreement Between Radiologists and Referring Clinicians on Follow-Up Imaging
November 18th 2024Agreement on follow-up imaging was 41 percent more likely with recommendations by thoracic radiologists and 36 percent less likely on recommendations for follow-up nuclear imaging, according to new research.