Photoacoustic Breast Imaging Can Reduce Skin Tone Biases While Detecting Cancer Early

Results from a study published in Biophotonics Discovery show which types of breast cancer imaging methods are less effective on patients with darker skin tones, which is an indicator of a bias. With these findings, the authors emphasize that next-generation imaging systems must consider all skin tones so screening will be more equitable.¹

Image credit: Valerii Apetroaiei | stock.adobe.com

Although breast cancer is a significant health concern worldwide, traditional imaging methods—such as a mammography—have limitations, particularly for women who have dense breast tissue. Early detection is crucial for effective and successful treatment, but skin tone bias may factor in to how quickly patients receive care. A promising alternative is photoacoustic imaging, which uses both light and sound to create detailed images of a patient’s breast tissue.¹

The authors investigated the impact of skin tone bias as a function of target sizes in 3 different photoacoustic image reconstruction methods: fast Fourier transform (FFT)-based reconstruction, delay-and-sum (DAS) beamforming, and short-lag spatial coherence (SLSC) beamforming. The 3 imaging methods were applied to channel data from multidomain simulations that had 757, 800, and 1064 wavelengths. Additionally, there were 11 target sizes (ranging from 0.5 to 3 nm), 18 different skin tones that ranged from very light to very dark (individual typology angles (ITAs) of –54 deg to 60 deg), as well as a previously validated realistic 3D breast model. Further, signal-to-noise ratio (SNR) and generalized contrast-to-noise ratio (gCNR) were also utilized to measure visibility.²

The study findings demonstrate that these traditional methods, FFT and DAS, do not properly visualize small targets under darker skin tones (ITA range: −54 deg to −33 deg), notably at 757 and 800 nm. Specifically, sizes that were 3 mm or smaller were poorly visible with lower mean SNR (≤2.38 to 4.21) and mean gCNR (≤0.46 to 0.74). Conversely, smaller targets in patients with lighter skin tones were generally more visible.²

Additionally, a 1064-nm transmit wavelength improved visualization with both DAS and SLSC beamforming, relative to both FFT reconstruction with a 1064 nm wavelength, and DAS or SLSC beamforming with 757 and 800 nm wavelengths. In combination with SLSC beamforming, the 1064-nm wavelength demonstrated the greatest improvements which enabled visualization of simulated target sizes that ranged from 0.5 to 3 mm underlying very light (ITA = 60 deg) to dark (ITA = −54 deg) skin tones. The mean SNR and gCNR were 10.01 or less and 1 or less, respectively. In addition, the authors noted that the visualization of simulated vessel structures derived from in vivo photoacoustic images was consistent with simulation-based target size expectations.²

Further, these findings highlight skin tone bias as a significant challenge in cancer screening methods. The authors note that the observed data are promising for the future of breast cancer diagnosis, and by addressing biases, photoacoustic imaging can potentially become a more reliable tool for early breast cancer detection and benefit women of all skin tones.¹

“This work was motivated by a previously poor understanding of photoacoustic imaging performance under combined variations of small target sizes and darker skin tones,” said senior and corresponding author Muyinatu Bell, PhD, in a news release. “Our results are enlightening, as we now have a better understanding of advanced photoacoustic imaging techniques and associated wavelengths necessary to detect small targets.”¹

REFERENCES

1. SPIE. Researchers shed light on skin tone bias in breast cancer imaging. News release. November 14, 2024. Accessed November 25, 2024. https://spie.org/news/researchers-shed-light-on-skin-tone-bias-in-breast-cancer-imaging#_=_

2. Rasquinha RD, Gubbi MR, Lediju Bell MA. Impact of skin tone on target size detectability in photoacoustic breast imaging. BIOS. 2024;2(1). doi:10.1117/1.BIOS.2.1.012502