Beamforming Reduces Photoacoustic Image Bias

Photoacoustic (PA) imaging offers a noninvasive means to obtain molecular and functional information about a variety of medical conditions. However, the quality of PA images, which are acquired when light illuminates the skin, is affected by the skin’s melanin content. The higher the amount of epidermal melanin, the more light is absorbed by the skin, increasing the chances of noise and artifacts.

To mitigate skin-tone bias in PA imaging, researchers at Johns Hopkins University and the University of São Paulo developed a technique to filter unwanted signals from the images of darker, more light-absorbent skin. By applying short-lag spatial coherence (SLSC) beamforming to PA imaging, the team found that it could acquire more accurate details about the location and presence of internal biological structures in dark-skinned individuals.

A comparison of two PA images of the same dark-skinned study participant shows that images made with conventional methods are cluttered, while the new imaging technique makes arteries easier to spot. Courtesy of Johns Hopkins University and the University of São Paulo, Brazil.

“When you’re imaging through skin with light, it’s kind of like the elephant in the room that there are important biases and challenges for people with darker skin compared to those with lighter skin tones,” professor Muyinatu Bell said.

Using PA data acquired at 750-, 810-, and 870-nm wavelengths from the forearms of 18 volunteers with various skin pigmentation levels, the researchers examined the impact of the epidermal melanin content on conventional, amplitude-based PA images. They also examined the capability of SLSC beamforming to reduce clutter artifact. They analyzed signal-to-noise ratio (SNR), clutter level, skin signal, and radial artery (RA) signal to provide quantitative characterizations of skin tone, clutter level, and image quality relationships for both conventional PA imaging and SLSC.

SLSC beamforming produced clear images of the arteries in the forearms of all study participants. In PA images made using conventional methods, it was almost impossible to distinguish the arteries in the darker-skinned participants.

The researchers demonstrated that PA signals from the skin were proportional to epidermal melanin content. As skin tone went from light to dark, light fluence throughout the imaging plane was reduced and clutter levels were increased, due to an increase in the skin’s optical absorption coefficient.

A comparison of the PA images from two different study participants, where imaging through darker skin shows more signal clutter, obscuring arteries, than imaging through light skin. Courtesy of Johns Hopkins University and the University of São Paulo, Brazil.

Conventional, amplitude-based PA images presented reasonable visualization of the RA for the lighter skin tones, with minimal clutter (e.g., −16 dB relative to the RA). However, clutter increased to −8 dB with the darker tones. Due to the high level of clutter, the researchers could not distinguish the RA from the background in the conventional PA images of the darker skin tones. This bias resulted in decreased SNR for darker tones.

When the researchers used SLSC beamforming, SNR improved for all the skin tones and wavelengths in the study. SLSC beamforming achieved a median SNR improvement of 3.8 dB, resulting in better RA visualization for all skin tones. The SLSC images of darker tones achieved an SNR comparable to that obtained when conventional PA imaging was used with light skin tones.

“We show that not only is there a problem with current methods but, more importantly, what we can do to reduce this bias,” Bell said. “Our work demonstrates that equitable imaging technology is possible.”

The researchers demonstrated that SLSC beamforming can mitigate skin-tone bias in PA imaging, resulting in images with good SNR and clear visualization of the RA, for a range of skin-tone categories. They are now working to apply their findings to breast cancer imaging, since blood vessels can accumulate in and around tumors. Bell believes the work of his team will improve surgical navigation, as well as medical diagnostics.

“We’re aiming to mitigate, and ideally eliminate, bias in imaging technologies by considering a wider diversity of people, whether it’s skin tones, breast densities, body mass indexes. These are currently outliers for standard imaging techniques,” Bell said. “Our goal is to maximize the capabilities of our imaging systems for a wider range of our patient population.”

The research was published in Photoacoustics (www.doi.org/10.1016/j.pacs.2023.100555).