A hand-held hybrid probe consists of a commercial US transducer located in the center and our NIR source detector fibers mounted in a housing, distributed at the periphery [16
]. The technical aspects of the NIR imager have been described in detail previously [14
]. Briefly, the imager consists of 12 pairs of dual-wavelength (780 and 830 nm) laser diodes, which are used as light sources, and their outputs are coupled to the probe through optical fibers. An additional wavelength (660 nm) is added to the system, which allows a reliable estimation of tumor oxygenation saturation. On the receiving side, eight photomultiplier tubes (PMTs) were used to detect diffusely scattered light from the tissue and eight optical fibers were used to couple detected light to the PMTs. The laser diodes' outputs were amplitude-modulated at 140 MHz and the detector outputs were demodulated to 20 kHz. The demodulated signals were further amplified and bandpass-filtered at 20 kHz. A reference signal of 20 kHz was also generated by directly mixing the detected radiofrequency (RF) signals with the RF signal generated from the oscillator. The reference signal was necessary for retrieving phase shifts. Eight detection signals and one reference were sampled and acquired into a computer simultaneously. The entire data acquisition took about 3 to 4 seconds, which was fast enough for acquiring data from patients.
The details of our dual-mesh optical imaging reconstruction algorithm have been described in Zhu et al. [15
]. Briefly, the NIR reconstruction takes advantages of US localization of lesions and segments the imaging volume into a finer grid in lesion region L and a coarser grid in nonlesion background region B. In all images, a 0.5 x 0.5 x 0.5-cm imaging grid was used for lesion and 1.5 x 1.5 x 1 cm was used for background region. A modified Born approximation is used to relate the scattered field Usd
measured at each source (s) and detector pair (d) to total absorption variations at wavelength in each volume element of two regions within the sample. The matrix form of image reconstruction is given by:
are weight matrices for lesion and background regions, respectively, and are calculated from background absorption and reduced scattering measurements acquired at the normal contralateral breast. ML
are total absorption distribution changes of lesion and background regions, respectively. The absorption distribution at each wavelength is obtained by dividing ML
by different voxel sizes in lesion and background tissue regions. With this dual-mesh scheme, the inversion is well conditioned and the image reconstruction converges in a few iterations.
Because the major chromophores are deoxygenated (deoxyHb) and oxygenated (oxyHb) hemoglobin in the wavelength range studied, we can estimate deoxyHb and oxyHb concentrations at each imaging voxel by inverting the following equations, voxel by voxel, as:
′) are absorption coefficients obtained at imaging voxel r
′, where wavelengths λ1
correspond to 780 and 830 nm in our system, respectively.
are extinction coefficients given in Cope [27
]. The total hemoglobin concentration totalHb(r
′) = deoxyHb(r
′) + oxyHb (r
′) and oxygenation saturation Y
100% can be calculated as:
. We have found that the best two wavelengths for total hemoglobin calculation are 780 and 830 nm, and the best two wavelengths for oxygen saturation calculation are 660 and 830 nm. These wavelength pairs were used in computing total hemoglobin concentration and oxygen saturation distributions reported in the Results section.
Because US resolution is less than 1 mm in depth for a typical 7.5-MHz US transducer such as the one we used, the boundaries between cancerous and normal tissue structures can be visualized well. However, spatial extensions of larger cancers, in general, are not well resolved in US. In addition, the NIR three-dimensional data are coregistered with one of the two spatial dimensions and depth dimensions of US. Another spatial dimension is estimated by assuming symmetry of lesion. Furthermore, the optical contrast may well extend beyond the tumor periphery due to angiogenesis development. Therefore, a larger region of interest (ROI), particularly in spatial dimensions than that visualized by US, is used for finer grid lesion region in the image reconstruction. The ROI used for each patient is listed in .
Histologic Microvessel Density Counts and Total Hemoglobin Measurements.
Clinical studies were performed at the University of Connecticut Health Center (UCHC). The UCHC IRB committee approved the human subject protocol. Written consents were obtained from all patients. Patients with palpable and non-palpable masses that were visible on clinical US and who were scheduled for biopsy or neoadjuvant chemotherapy were enrolled as research subjects. Six patients with tumors ranging from 2.5 to 4 cm were studied. For each patient, US images and optical measurements were acquired simultaneously at multiple locations including the lesion region, a normal region of the same breast if the breast was large, and a normal symmetric region of the contralateral breast. The optical data acquired at normal region with the best linear amplitude and phase profiles were used as reference for calculating the scattered field caused by lesions. The total hemoglobin concentration maps are quantified by measuring the maximum value at each depth (layer) and the average within 50% of the maximum value. Because the hand-held probe can be easily rotated or translated, at least three coregistered US and NIR data sets were acquired at the lesion location, and the corresponding optical absorption maps as well as the total hemoglobin concentration distribution were reconstructed using the coregistered US. The data given in , column 5 (left column) are average values obtained from at least three sets of NIR images.
To correlate the imaged hemoglobin distribution with histology microvessel density, we have performed microvessel density counts. Samples obtained at biopsy or definitive surgery were used for counting. For each sample, sections 3 to 5 µm thick were stained on an immunohistochemistry slide staining system (DAKO autostainer) with factor 8/86 mouse monoclonal antibody (antihuman von Willebrand factor; DAKO Corp., Carpinteria, CA) at 1:100 dilution digested by proteinase K for 3 minutes by labeled polymers (DAKO EnVision plus) using the immunoperoxidase method. Histologic microvessel density count was assessed by immunohistochemistry as initially proposed by Weidner et al. [20
]. The microvessel density counts were performed in 10 consecutive fields with the use of an ocular grid at x200 magnification. The first field chosen was a hotspot (area of maximum vascular density either within the infiltrating tumor mass or at the tumor-stromal interface). Because the cancers were large, two to three separate sample blocks were selected for microvessel counts as per specimen orientation in the surgical pathology report. The performer was blinded from the NIR imaging results.
The surgeon always orients the excised specimen with sutures designating the resection margins. Tumor sampling for histologic studies demonstrates the relationship of the tumor to the designated resection margins. Because the patients were imaged from the anterior approach with the patient in supine position, orientation of the anterior and deep (posterior) locations within the optical and US images could be easily linked to the orientation of the surgical material.
The linear regression curve of microvessel density count versus
measured maximum total hemoglobin concentration at the corresponding location was obtained from least square solutions. The correlation coefficient that reflecting the goodness of fit was computed as:
is a sample point of microvessel density count and its corresponding maximum total hemoglobin concentration, and
is the mean value of the corresponding variables. The statistical significance was tested on correlation coefficients with a confidence level of 0.05.