The pathological evaluation of breast cancer after neoadjuvant therapy requires accurate specimen sampling so that essential information can be obtained to measure the maximum diameter for RCB assessment. This evaluation system contains 6 elementary factors including two maximum diameters of the largest cross-section of tumor bed, overall cancer cellularity, percentage of cancer area, number of positive lymph nodes, and diameter of largest metastasis. A precise sampling of the largest cross-section will enable us to gain accurate data on four elements. Thus, the current standard for RCB evaluation requires pathologists to accurately identify the various tissue types, e.g., tumors, fibers, fats, in the tumor bed[30, 31] .
Conventional methods of specimen sampling mainly rely on visual observation, and the conclusion is mostly based on experience. For example, we consider gray-yellow areas as fat, gray areas as fiber, and gray-white areas as tumor. However, the carcinoma is very heterogenous; hence these evaluation are subjective. Traditional methods also include touching to estimate the sample’s texture, which has poor reproducibility. Moreover, lesions inside the specimen are not visible by eyes. If the pathologist fails to accurately determine the distribution of tumors, there are at least two severe consequences: the blind sampling, which results in an increased workload, and a damage to the tumor bed of the largest cross-section and a reduced accuracy. Other methods such as X-ray imaging have been used, but they are expensive with ionizing radiation and not showing high contrasts among fats, fibers, and tumors.
In recent years, advancement in technologies has improved the clinical diagnosis. For instance, hyperspectral and multispectral provide certain resolution to human tissues[32–37], and radiation can be replaced by visible light (400-700nm) in some area[38–40]. Significantly, we developed the high dynamic range dual-modal white light imaging system for tumor bed sampling.
The HDWIS system we designed contains three primary techniques: transmissive imaging, polarized imaging, and high dynamic range (HDR) imaging, which are already widely used in photography, industrial inspection, and certain medical fields. For example, transmissive imaging is used for egg sorting and grading . Polarized imaging is widely used for industrial material classification  and skin examination . HDR  is also a popular technology in the field of photography and exists in almost all current DSLR cameras and mobile phones to expand the range of both contrast and color of the image.
The HDWIS imaging system is able to produce anti-glaring reflection and HDR transmission specimen images and provide pathologists with information not only from surface but also inside the specimen. The multi-exposure transmission images are also particularly important when the tumor beds have larger thickness value. Based on the difference in transmittance, we can distinguish tissue types such as tumors, fibers, and fats. Although the transmittance of all tissue types declines with increasing thickness, there is no linear correlation between thickness and transmittance as shown in Fig. 3B, suggesting that the transmittance of tissue is affected not only by the thickness but also by the characteristics/texture of the tissues. By combining the transmittance information and the texture features of the transmission images, we might identify the locations of fibers, tumors, and fats more confidently.
In the transmission images by HDWIS, the fibrosis areas of the tumor beds displayed directional texture in 61% of the cases, while the textures of tumor areas were disordered, non-directional or invisible. Under the microscope, we observed that the fibroblasts displayed directional arrangement, while cancer cells were irregular (solid, sieve, or scattered). These micro-distribution patterns determined the tissue appearance and types, which displayed clear boundaries in the transmission images due to the differences in transmittance. The HDWIS reflection images are also important in this analysis. HDWIS with anti-reflection and ultra-fine focal-length-change function could further improve the assessment accuracy. The anti-reflection function could assist us in avoiding the omission of granular areas (DCIS), and the high-resolution camera could enlarge the texture images of small lesions for easy visual identification. However, for resolving power (R.P.), reflection image with high resolution and the visual observation give similar results. When the fibers are formed by hyaline, the directional texture disappears. Moreover, if the cancer cells are scarce, and fibers account for a large proportion of the tumor bed, tumors and fibers could not be differentiated through transmission images. By combining information from tissue colors, textures, and zoomed-in details of anti-reflection images, we will be able to overcome the complexity caused by tumor heterogeneity.
HDWIS has the highest detection sensitivity for fibers and tumors compared to the traditional methods. The transmission imaging method can identify the largest tumor bed area of 3–7 mm. HDR can further expand the dynamic range of the transmission images and provide more colors and details in both bright and dark areas. Additionally, the HDR transmission images could help find the fibrosis areas which were hard to detect by other techniques including the original 2D WSI and X-ray. The deep WSI and recalculated specificity shown in Fig. 6C1-H2 confirmed our observation. Although the imaging principles of X-ray and transmitted light are similar, the radiation dose of the cabinet X-ray system is fixed such that some samples cannot be examined in detail. HDWIS solves this issue by relying on multiple exposure transmission images and the optimized synthetic HDRs. Hence, HDWIS is superior in examining thick specimens via showing the tumor bed deep inside the specimen.
Nevertheless, HDWIS may not work well for tumor beds with complex non-concentric contraction patterns[45–47] or small lesions that are only detectable by histological microscope. The study scope of this work is confined to excised breast cancer specimens. Future studies may extend to other cancer specimens. Although the collected specimens are formalin-fixed, future studies may also extend to fresh cancer specimens as well by using the proposed method as a potential intraoperative cancer margin assessment tool. In addition, the ability of HDWIS to identify necrosis, inflammation, and other components requires further investigation.