In recent decades, various techniques have been adopted to improve NIRST system performance, which can facilitate the use of NIRST in breast cancer detection, diagnosis, and treatment. The purpose of this study is to review the current progress on NIRST systems and summarize their advantages and limitations. We also report recent clinical applications of NIRST systems in breast imaging and discuss the challenges and future developments.

Breast cancer is the most common cancer diagnosed among women worldwide, which accounts for 11.7% of all new cancer diagnoses in 2020. Breast cancer mortality rates decrease significantly when breast tumor is detected early using imaging tools. As an emerging imaging technique, near-infrared spectral tomography (NIRST) has demonstrated potential in breast imaging owing to its nonionizing radiation and high sensitivity and cost-effectiveness. The aim of NIRST is to resolve three-dimensional images of tissue optical properties and chromophore concentrations from acquired multi-wavelength measurements. Therefore, functional information related to biological tissue can be obtained, which is indistinguishable using current clinical breast-imaging modalities. However, NIRST exhibits poor spatial resolution because of light scattering in biological tissues. NIRST system is the key ingredient for producing NIRST images of high spatial resolution.

This paper presents a review of imaging types (Fig. 2) involved in data acquisition. First, continuous wave (CW) systems, including available commercial instruments, are introduced (Fig. 3). The widely used frequency-domain (FD) and time-domain (TD) systems are summarized (Figs. 4 and 5). The emerging hybrid imaging types and relevant prototype systems are also reviewed (Fig. 6). The integration of conventional breast cancer-imaging systems into NIRST can enhance spatial resolution of NIRST and improve lesion characterization. Therefore, the multimodality imaging systems widely used in breast imaging are also reported (Figs. 7 and 8), particularly in magnetic resonance imaging (MRI)/NIRST interfaces. As incorporating structural information is critical for the accurate clinical diagnosis of breast cancer, the methods including hard prior, soft prior, direct regularization imaging, and the new deep learning methods are discussed (Fig. 9). Their applications in breast cancer diagnosis and prediction response to breast cancer neoadjuvant chemotherapy are also demonstrated (Figs. 10 and 11).

Near-infrared spectral tomography can provide functional information regarding breast tissue and be used as a supplemental imaging tool for clinical breast cancer-imaging modalities. However, the primary restriction of NIRST is poor spatial resolution. Recent developments in hybrid imaging types and multimodality imaging have facilitated studies on breast cancer management. In addition, deep learning has been applied to NIRST to improve lesion characterization and reduce computational time. The proposed method is expected to assist in breast diagnosis.