Added Intelligence Transforms Medical Sensors Into Diagnostic Devices

MARIE FREEBODY, CONTRIBUTING EDITOR, marie.freebody@photonics.com

As sensors shrink in size, they are able to reach places that were previously inaccessible — such as inside the human body. What is more, incorporating such increasingly tiny sensors within a medical device means there is more space left for added functionality. Medical devices of the future will not only have the capacity to sense but will also perform meaningful analysis and diagnostics.

An aging and expanding population is accelerating the development of new and different types of medical equipment, including various sensors used inside both equipment and patients’ bodies.

According to a 2017 market report¹ by market research and strategic consulting company Yole Développement of Lyon, France, the medical industry has a growing interest in solid-state technologies in order to answer the challenges of miniaturization, patient safety, early diagnostics, low power consumption and cost-savings.

“Three hundred fifty million dollars of solid-state optical sensor devices for medical imaging applications has been sold in 2016,” said Benjamin Roussel, business unit manager of Yole’s MedTech activity. Yole expects a growth of 8.3 percent in the next five years, he said.

Today’s optical sensors measure a variety of biometrics, including heart rate and blood oxygen concentration and perfusion. Each of these measurements involves illuminating a light source at a specific wavelength into the body through the skin. This light scatters and modulates based on the properties and characteristics of the target tissue through which it travels.

For example, light illuminated across a finger will scatter and be absorbed due to the blood flowing through that area. If one can detect the parameters of the propagating light, a waveform — called a photoplethysmogram (PPG) — can be constructed, which correlates strongly with a subject’s heart rate.

In the case of on-body sensors used in wearable devices (see “Wearable Health Care” sidebar below), pulse oximetry is the most established optical monitoring technique. Dating back to the 1970s, it tracks heart rate and the level of arterial blood oxygen saturation (SpO₂). In recent years, this has been extended to measure other blood components, including carboxyhemoglobin and methemoglobin.

“A somewhat similar technique based on near-infrared spectroscopy is being used increasingly for neonatal cerebral oxygen monitoring,” said Wim Verkruysse, principal scientist at Philips Research, Eindhoven, the Netherlands. “Availability of low-cost, high-resolution, visible and/or near-infrared light imaging sensors drives the development of their medical applications.”

One common example can be found at most major airports in the form of IR sensors used for temperature estimation in fever scanners.

With only a few camera-based products available on the market, which focus almost exclusively on monitoring mechanical body properties such as respiratory rate, Philips’ research scientists believe camera-based devices will provide exciting new opportunities for the medical sensor industry.

“Due to their advantages, cameras constitute promising future replacements of contact sensors,” said Marek Bartula, research scientist at Philips. “Many advantages, including convenience, lower infection rates, elimination of consumable costs, potentially increased robustness and multimodality of camera-based solutions, set it as one of the most promising directions in the future of patient monitoring.”

In cases where no convenient on-body solution exists or the application does not justify the cost of consumables, contactless measurement enables previously unmonitored patients to be observed.

“In principle, any on-body optical measurement could be potentially performed in a contactless way,” Bartula said. “Philips [is] the first in the world [to demonstrate] that the absolute SpO₂ can be accurately monitored across multiple patients.”

The practical challenge facing camera makers lies in the often orders of magnitude lower signal-to-noise ratios caused by either a smaller signal or a greater sensitivity to noise sources such as patient motion or ambient light.

“This can be largely compensated by heavy oversampling with thousands or even millions of photodetectors, instead of single elements placed on skin or against other tissues, and smart algorithms that make use of this oversampling,” Bartula said. “In general, the noncontact nature allows for better, unperturbed measurement and even increased motion robustness when compared with contact probes. Philips’ current algorithms for noncontact, camera-based heart rate monitoring during subject movements already demonstrate this.”

Endoscopic pill cameras can be used for small bowel imaging.

Yole’s Roussel also notes the growing trend of miniaturized cameras coming from the consumer market, which have supported the emergence of new applications in endoscopy. Examples include minimally invasive procedures through miniaturized endoscopes, small bowel imaging with camera pills, and disposable flexible endoscopes in colonoscopy or bronchoscopy to avoid cross-contamination.

Ophthalmic diagnostics

In ophthalmic diagnostics, medical optical sensors can measure the shape and integrity of the cornea, crystalline lens, and distances between refractive surfaces of the eye from the retina, and measure structural properties of the retina and specific structures within it.

Over the past decade, optical coherence tomography (OCT) has become the fastest-moving imaging technology in ophthalmology. For example, instruments such as the Zeiss IOLMaster 700 can precisely measure distances between various optical elements of the eye, which helps determine the optimum intraocular lens power needed for cataract surgery.

Fourier Domain OCT also is used to image and measure tissues in the front and back of the eye; such tissue measurements have been reduced to standard clinical metrics that aid in diagnosing and managing diseases such as diabetic retinopathy, macular degeneration and glaucoma.

Zeiss IOLMaster 700 full-length OCT image of the eye. Courtesy of Prof. W. Sekundo, Philipps University Hospital Marburg, Germany, and Carl Zeiss Meditec.

Recently, OCT Angiography (OCT-A) has been approved for use in ophthalmology. Unlike traditional fluorescein angiography in which a contrast agent is injected into the bloodstream, OCT-A detects the movement of blood and tissues and produces extremely high-resolution images of blood vessels in the retina, the choroid, the optic nerve and the conjunctiva.

One example of an OCT-A instrument is the Zeiss AngioPlex, which has strong potential for use in diabetic retinopathy, age-related macular degeneration, glaucoma and other blinding eye diseases.

A continuing challenge is acquiring larger OCT and OCT-A scans to be able to view more of the retina. But OCT and OCT-A could challenge clinics and ophthalmologists with the data to be managed, in the same way that MRI and CAT scanners have historically challenged radiologists.

This has provided significant opportunities for OCT manufacturers to help eye doctors analyze OCT and OCT-A images and data with additional opportunities emerging in the area of data transmission and storage.

Multispectral sensors

Another trend is the application of multispectral sensors to health care. Multispectral sensors make it easier to distinguish various compounds and assess blood and tissue composition. Their increasing availability is down to the many cost-effective applications originally developed for use in the agricultural and food industries.

Many environmental and agricultural secrets have been unlocked thanks to hyperspectral sensors such as those developed by hyperspectral imaging specialist Headwall Photonics in Bolton, Mass. The medical market is exploring the potential of applying similar devices to image tissues at microscopic or cellular levels.

Such noninvasive, noncontact and nonionizing analysis can provide accurate spectral information relating to a patient, a tissue sample, or a disease condition. Instead of a single image, a data cube is obtained — a stack of images collected at different wavelengths with each pixel containing an entire spectrum of data.

“For example, the HELICoiD [HypErspectraL Imaging Cancer Detection] project in Europe partnered very successfully with Headwall to produce a hyperspectral-based system capable of detecting cancerous brain cells with a very high degree of spectral and spatial resolution,” said David Bannon, CEO at Headwall Photonics. “In other words, being able to detect what it is, and where it is.”

The challenge with cancerous brain cells in particular is that they are exceptionally difficult to discern from normal brain cells. The HELICoiD project, which finished at the end of last year, involved a collaboration of four universities, three industrial partners and two hospitals, resulting in a methodology to discriminate between healthy and malignant tissues in real-time during surgery.

“In other types of cancer, the distinction from normal cells is a bit easier to detect. But brain cancer is different; the HELICoiD Project proved that hyperspectral imaging sensors are able to distinguish good cells from bad,” Bannon said. “This is particularly meaningful because brain cells do not regenerate; the surgeon needs a very precise map before he operates, and hyperspectral image data provides him with that map.”

A new approach for real-time human brain cancer detection involves development of spatial-spectral cancer detection algorithms and spectral signatures in conjunction with Headwall’s hyperspectral imaging sensor. Courtesy of HELICoiD project.

Deep learning is used by applying an algorithm to identify cancerous and healthy tissue in hyperspectral images acquired during surgery [see “Deep Learning: Google DeepMind Identifies Skin Cancer” sidebar]. In the future, this information could be provided to the surgeon in real time by overlaying conventional images with a color-map that indicates the likelihood of that particular area being cancerous.

From sensing to imaging, medical optical sensors are highly segmented in terms of technologies and applications. But one thing is for certain: As optical components continue to reduce in size and cost, exciting developments are on the horizon.

“It’s amazing to see how disruptive technologies at the sensor level can lead to new applications and modify the industry landscape,” Roussel said.

References

1. Yole Développement (January 2017). Solid-State Medical Imaging 2017, www.i-micronews.com/category-listing/product/solid-state-medical-imaging-2017.html.

2. A. Esteva et al. (February 2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, Vol. 542, pp. 115-118, doi:10.1038/nature21056.

Deep Learning: Google Deep Mind Identifies Skin Cancer

Researchers at Stanford University have developed a new system that uses artificial intelligence to classify skin cancers as successfully as human experts. This could help early detection and demonstrates the growing interest in applying algorithms to medical applications.

Based on image-recognition in conjunction with a deep learning algorithm built by Google, which had already been presented with 1.28 million images of objects such as cats, dogs and cups, the Stanford team fed the system with more than 127,000 clinical images of skin lesions encompassing many different skin diseases.

Their approach, described in Nature², could be developed for smartphones, increasing access to screening and providing a low-cost way of checking whether skin lesions are cause for concern.

Wearable Health Care

In wearable electronics, the majority of optical sensors use light-emitting diodes (LEDs) and photodetectors to measure changes in blood hemodynamics and vasculature metrics. Examples of this are currently found in smartwatches that use green LEDs and appropriate photodetectors to measure heart rate at the wrist.

Smartwatches use green LEDs and appropriate photodetectors to measure heart rate at the wrist.

In medical settings, a similar technology is used to measure blood oxygen concentrations — SpO₂ — using devices called pulse oximeters. These devices use red and infrared LEDs with photodetectors to resolve the relative concentrations of oxygenated and deoxygenated hemoglobin cells in the blood stream.

The pulse oximeter is the most established optical monitoring technique for on-body sensors in wearables, tracking heart rate and arterial blood oxygen saturation.

In mobile and wearable applications, the mechanical interface with skin is critical. New battery technologies and flexible/stretchable electronics technologies are beginning to mature from research labs to commercial markets, addressing these important challenges related to energy densities, size and mechanics.

Optical sensors and other sensing modalities have driven rapid advances in digital health in recent years.

“With motion-based sensors — in other words, accelerometers — it is now possible to contextualize the data streams from the optical sensors,” said Milan Raj, director of Advanced Technology at wearable biosensor specialist MC10 in Lexington, Mass. “We can thus infer why your optical heart rate measurement is elevated, if we know your activity and stress levels. This type of sensor fusion could drive novel algorithms.”