Open Access
Add to BibSonomy
Abstract
Integration of magnetic resonance imaging (MRI) and near-infrared spectral tomography (NIRST) has yielded promising diagnostic performance for breast imaging in the past. This study focused on whether MRI-guided NIRST can quantify hemoglobin concentration using only continuous wave (CW) measurements. Patients were classified into four breast density groups based on their MRIs. Optical scattering properties were assigned based on average values obtained from these density groups, and MRI-guided NIRST images were reconstructed from calibrated CW data. Total hemoglobin (HbT) contrast between suspected lesions and surrounding normal tissue was used as an indicator of the malignancy. Results obtained from simulations and twenty-four patient cases indicate that the diagnostic power when using only CW data to differentiate malignant from benign abnormalities is similar to that obtained from combined frequency domain (FD) and CW data. These findings suggest that eliminating FD detection to reduce the cost and complexity of MRI-guided NIRST is possible.
© 2021 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
near-infrared spectral tomography (NIRST) has been investigated as an adjuvant breast imaging modality [1–3]. Compared to the low sensitivity of ultrasound [4] and the risk of exposure to ionizing radiation of mammograms or digital tomosynthesis [5], it is non-ionizing, fast, relatively inexpensive, poses little, if any, health hazards, and therefore, can be used repeatedly to image subjects without safety concerns [6]. When NIRST is used to acquire volumetric measurements by placing sources and detectors around the breast, measurements in a plane of diameter up to 106 mm have been detected [7]. Studies have demonstrated that NIRST can differentiate benign from malignant breast abnormalities [8–14], and monitor the pathological response of breast cancer to neoadjuvant chemotherapy (NAC) [15,16].
NIRST signal measurement techniques for breast imaging have involved 3 types of methods [17–19]: continuous wave (CW) [20], frequency domain (FD) [21], and time domain (TD) [22]. CW imaging systems detect light intensity. Accordingly, they are fast, low cost and very compatible with MRI, but they do not sense signal phase which is associated strongly with tissue optical scattering. FD imaging systems use light sources which are intensity-modulated at high temporal frequency (∼ 100 MHz) to illuminate the breast and acquire light amplitude and phase shift [23]. Aa a result, FD imaging requires several additional system components that are not only expensive but also more complex relative to its CW counterpart. For example, source light has to be intensity modulated with frequencies in the 100MHz range to obtain adequate phase shift, which requires a laser sub-source system to modulate the light intensity accurately with a set of microwave devices (∼${\$}$10K). For signal detection, a microwave demodulation subsystem, which involves low-noise high frequency amplifiers and heterodyning detection modules, is required. Further, since the optical detectors need very high sensitivity and fast time response sufficient to detect phase changes in very weak (∼pW) optical signals (transmitted through breast tissue), photomultiplier tubes (PMTs, ∼${\$}$1K/channel) are needed relative to photodiodes (for CW detection, ∼${\$}$10/channel). PMTs are not MRI-compatible; therefore, bulky, heavy, long fiber bundles (e.g., 12 meter-long, 10mm-in diameter, ${\$}$60 K for 16 channels as in [9]) are the only solution to transfer diffuse light collected at the breast surface to PMTs located on an instrumentation cart placed in the MR console room (whereas photodiodes can be put in the MRI scanner). Space is limited by the MRI breast coil and scanner, itself, to the extent that only about 16 of these large fiber bundles can be used to cover the breast surface. Tumor coverage has been a key problem with MR-guided NIRST when based on FD data acquisition (our pilot study showed that ∼30% of exams resulted in low optical data sensitivity to tumor because of inadequate fiber placement and/or breast coverage [24]). CW MRI-guided NIRST systems can have many more channels, for example up to 64 CW channels or more per breast, because of their low cost and compact size. Thus, FD instruments are more complex and costly.
In contrast to CW imaging, chromophores related to absorption and scattering can be estimated simultaneously with FD imaging [23]. Studies have shown that the quantitative accuracy of recovered NIRST parameters is improved when using FD techniques compared to CW methods [25,26]. TD systems apply short laser pulses with temporal spread below a nanosecond, and detect the broadened spread of the pulse exiting the breast surface. They also estimate breast absorption and scattering coefficients, but are even more expensive and more difficult to implement in a clinical setting relative to FD approaches.
Recently, several NIRST studies have combined FD and CW (FD/CW) systems for breast imaging [18,27], in which scattering amplitude (SA) and scattering power (SP) are first calibrated and recovered from FD phase data, and the distribution of hemoglobin concentration and blood oxygen saturation are then reconstructed from FD/CW amplitude data based on calibrated SA and SP values. Wang et al. demonstrated that quantitative accuracy in reconstructed NIRST images is improved with FD/CW data relative to FD data alone [19]. Diagnostic sensitivity, specificity and AUC reach 90%, 100% and 0.94, respectively with tissue optical index (TOI) as the NIRST biomarker [9]. However, FD/CW systems increase acquisition time and operational complexity/cost.
Hemoglobin concentration appears to be the fundamental biomarker for breast diagnosis and NAC monitoring [28–30]. SA and SP have shown inconsistent differentiation of breast tissue types. Spinelli et al. showed no significant differentiation [31], while Cerussi et al. found increased SP in cancers [32]. As a result, CW systems may be the best choice for enhancing breast diagnosis through MRI-guided NIRST, especially if SA and SP of the breast can be pre-determined from its density. Here, MRI-based assessment of breast density could be used to guide the choice of SA and SP values, since optical scattering parameters are correlated with breast density: fatty breast have low scattering power and extremely dense breasts have high values [33]. Furthermore, MRI data can be incorporated into NIRST to overcome its poor spatial resolution and ill-posedness without image segmentation [10,11,13].
In this paper, we investigate whether NIRST, when combined with MRI as a simultaneous, co-registered breast exam, MRI-guided NIRST, can quantify hemoglobin concentration using only continuous wave (CW) measurements. The approach assigns optical scattering parameters based on MRI-derived breast density values, and then reconstructs MRI-guided NIRST images from calibrated CW data. Results from simulations and twenty-four patient cases were analyzed to estimate the relative diagnostic performance of the new CW MRI-guided NIRST breast imaging approach. Findings from the study are important because they will determine whether eliminating FD detection to reduce cost and complexity of MRI-guided NIRST systems is possible.
2. Methods
2.1 Imaging procedures
A multi-modality MRI-NIRST system developed at Dartmouth College was used to acquire NIRST data and MRI images simultaneously from the breasts of women with undiagnosed abnormalities at time of the imaging exam. Technical details about the combined system can be found in previous publications [9,14,24].
MRI acquisition consisted of standard (T1, T2, DWI) and dynamic contrast enhanced (DCE) sequences. NIRST volumetric data (from 16-located sources/detectors) were collected at 6 shorter wavelengths (661, 735, 785, 808, 826 and 852 nm) for FD measurements and 3 longer wavelengths (903, 912 and 948 nm) for CW measurements. Amplitude data from all nine wavelengths were used to estimate chromophore concentrations related to breast tissue absorbers.
2.2 Subjects
A MRI/NIRST imaging protocol was approved by the Institutional Review Board (IRB) for the Committee for the Protection of Human Subjects at Dartmouth College and at Xijing Hospital. Criteria for patient enrollment were (i) age greater than 18 years, (ii) breast size and epithelial integrity adequate to allow NIRST data acquisition, (iii) an undiagnosed palpable mass or confirmed lesion from clinical imaging, (iv) no prior biopsy within the previous 10 days, (v) ability to provide informed written consent, and (vi) no serious associated psychiatric illnesses [9]. The original study collected DCE MRI-guided NIRST exam data from 44 women meeting these inclusion criteria. Due to optical fiber bundle diameter, length and mass, only 16 channels were available for FD measurements in the MRI bore with the NIRST instrumentation, and detector coverage of the breast abnormality of interest with optical sensitivity > 2% (which has been determined to be the minimum threshold for acceptable NIRST image reconstruction [24]) occurred in the 24 patients evaluated here. Patient information is summarized in Table 1.
Table 1. Clinical Information of Patients and NIRST Scattering Properties Used in the Study.
Based on MRI T1 images, breast density of each participant was assessed by an experienced breast radiologist and assigned to one of four subgroups: 1 fatty (F), 10 scattered (S), 6 heterogeneously dense (HD), and 7 extremely dense (ED). Body mass index (BMI) was also assessed, and averaged across breast density categories for the subjects enrolled (see Table 1).
2.3 Image reconstruction
Figure 1 shows the flowchart for NIRST image reconstruction. First, a 3D finite element (FE) mesh was generated from MRI T1 images through the open source software NIRFASTSlicer [34]. To minimize systematic errors from variations in laser intensity at different wavelengths, light coupling efficiency at different fiber positions, and detector sensitivity, measured amplitude data for each source-detector pair were calibrated with a homogeneous reference data set acquired immediately prior to a patient exam. In this study, a direct calibration method is developed for data calibration, instead of the popular slope-dependent calibration approach [35,36]. The direct calibration algorithm is described as follows: calibrated data (amplitude only), ${f_c}$, for image reconstruction was formed from patient measurements, ${d_p}$, recorded with the imaging instrument through normalization by
where ${d_r}$ is measured on a homogeneous reference phantom and ${f_r}$ is the corresponding simulated data on the FE mesh. This formula corrected for errors in fiber coupling that were presented in the data acquired from the reference phantom and patient, and fitted all of the data to the model in a single step.Average optical scattering parameters (SA and SP) used in experiments described in this paper are summarized in Table 1. Here, SA and SP were obtained from prior data [9] in which subjects were grouped into 4 breast density categories and their scattering parameter values were averaged to produce the entries in Table 1.
Initial estimates of chromophore concentrations for HbO, Hb, water and lipid for NIRST reconstruction were found by fitting amplitude data [35,36]. Absorption coefficients were determined at each wavelength using a hard-prior based reconstruction [37,38] in which the breast was assumed to be a single optically homogenous region. This method fits the data to the model iteratively, is independent of source strength, and can be used with a small number of data points. Initial property estimates were fed into a single-region, multi-wavelength spectral reconstruction in order to impose spectral constraints during the fitting process which avoids local minima [19].
After the initial homogeneous property estimates were generated, NIRST images were recovered by minimizing the difference between calibrated data, ${f_c}$, and model data, ${f_m}$, obtained by solving diffusion equation:
where x contains the chromophore concentrations for HbO, Hb, water and lipids.MRI DCE images provided anatomical information for NIRST reconstruction. To encode MRI DCE data, a direct regularization imaging (DRI) method was adopted [10,13,39]. Accordingly, the minimization in Eq. (2) was augmented to become
Based on the Levenberg-Marquardt (LM) procedure, an update to the chromophore concentrations, $\varDelta {x_k}$, at kth iteration was obtained from Eq. (3) as
where ${J_k}$ is the Jacobian matrix at the kth iteration, which has the formFurther details on constructing the Jacobian matrix ${J_{c,{\lambda _n}}}({c = HbO,Hb,water,lipid} )$ at each wavelength ${\lambda _n}(n = 1,2, \ldots ,9)$ can be found in [34]. ${\beta _k}$ is the regularization parameter at the kth iteration, and is set to be ${\beta _k} = 10\ast \max (diag(J_k^T{J_k}))$, ${f_m}({x_{k - 1}})$ is the forward solution using the estimated parameters ${x_{k - 1}}$ from the $(k - 1)\textrm{th}$ iteration. The DRI-based NIRST reconstruction was implemented by modifying NIRFAST [34].
From the recovered chromophore concentrations, total hemoglobin (HbT) contrast (the ratio of average HbT in the abnormal region to surrounding normal tissue) was calculated and used to assess differences in malignant and benign abnormalities in subsequent analyses.
2.4 Numerical simulation
Three-dimensional (3D) simulations were performed on patient-specific breast MRIs to evaluate the performance of MRI-guided NIRST image reconstruction based on CW amplitude data alone. Here, MR scans were segmented into 3 different tissue regions (adipose, fibroglandular, and tumor tissues) in order to assign optical property values for generating synthetic data. Two simulation studies were generated: Case 1 was taken from an extremely dense breast with a 23 mm × 40 mm × 70 mm malignancy in the lower middle quadrant approximately 3 mm below the surface; Case 2 originated from a breast with scattered density and a cancer positioned centrally 8 mm from the skin surface. The lesion size is 35 mm × 46 mm × 74 mm. After image segmentation, HbT, water, lipids, SA and SP values for tumor, fibroglandular and adipose tissues were assigned to each region for numerical simulations [9,13,33]. Data were generated synthetically from 16 co-located source-detector positions placed around the breast. For each source illumination, measurements were collected at 15 detector locations for 9 different wavelengths leading to a total of 2160 values (16 sources × 15 detectors × 9 wavelengths) with 5% added random gaussian noise. To compare outcomes from CW data with assigned SA and SP parameters, NIRST reconstructions were also computed with FD/CW data – amplitude and phase at six short wavelengths and amplitude at the remaining three longer wavelengths – to recover HbT, StO2, water, lipids, as well as SA and SP images, where ${J_k}$ in Eq. (6) was calculated according to a previous study [18].
2.5 Statistical analysis
Statistical analysis was carried out using Matlab 2018b. A student's t-test was performed to examine differences between average HbT contrast in malignant and benign breast abnormalities. Receiver-operating characteristic (ROC) analysis was also performed to estimate sensitivity and specificity for detection of malignant versus benign breast abnormalities, and to evaluate differences in the discrimination of malignant versus benign lesions as a function of cut-off value across the parameter space. Significance for all statistical tests was assumed at a confidence interval of 95% (P<0.05) for a two-tailed distribution. We identified the best cutoff point corresponding to the largest summation of the average sensitivity and specificity in the ROC curve.
3. Results
3.1 Simulation data
Figure 2(a) and 2(b) show data from MRI T1 and DCE images for simulation Case 1. Typical HbT, water, lipids, SA and SP values for tumor, adipose and fibroglandular tissues (see Table 2) were assigned to each region [9,13,33]. True HbT images appear in Fig. 2(c). Values of SA and SP used in CW reconstruction were selected from the extremely dense group in Table 1. Average SA and SP values estimated from FD/CW data in tumor were 1.19 and 0.84, while those used in CW data reconstruction were 1.26 and 0.36, respectively. Figure 2(d) and 2(e) show images recovered from FD/CW and CW data, respectively. Quantitative comparisons of the estimated HbT, and the error in HbT (relative to true values) are also reported in Table 2. As shown in Fig. 2(d) and (e) and Table 2, errors in HbT values recovered with FD/CW data ranged from 9.2% to 76.9% relative to their true values; while they were 4.3% to 79.3% when only CW data were used. HbT contrast between tumor and surrounding normal tissue was higher with CW data alone (about 1.8) relative to FD/CW data (about 1.5).
Fig. 2. Images from simulation Case 1: extremely dense breast. (a) MRI T1; (b) MRI DCE; (c) Ground truth NIRST image overlays; (d) HbT images reconstructed from FD/CW data; (e) Same as (d) with CW data only. NIRST images are overlaid on the T1 MRI cross-section. The blue lines denote tumor locations in (b).
Table 2. True and reconstructed NIRST chromophore concentrations in simulation Case 1.
MRI T1 and DCE images from simulation Case 2 are shown in Fig. 3(a) and 3(b), respectively. Reconstructed images with FD/CW and CW data appear in Fig. 3(d) and 3(e), respectively, and can be compared to ground truth (in Fig. 3(c)). Quantitative reconstruction results are compiled in Table 3. Similarly to simulation Case 1, errors in recovered HbT values (relative to true values in each region) varied, but HbT contrast between tumor and surrounding normal tissue was higher with CW data alone (about 3.6) relative to FD/CW data (2.5) used for image reconstruction. Results also indicate that when scattering parameters are assigned based on breast density, recovered HbT contrast between the tumor and the surrounding tissues is higher with CW data relative to FD/CW data.
Fig. 3. Images from simulation Case 2: scattered breast density. (a) MRI T1; (b) MRI DCE; (c) Ground truth NIRST image overlays; (d) HbT images reconstructed from FD/CW data; (e) Same as (d) for CW data only. NIRST images are overlaid on the T1 MRI cross-section. The blue lines denote tumor locations in (b).
Table 3. True and reconstructed NIRST chromophore concentrations in simulation Case 2.
Errors in recovered HbT concentration due to mis-assigned breast density were studied. As shown in Fig. 4(a), when extremely dense breast density (purple) in the first simulation case (Fig. 2) was assigned to each of fatty (blue), scattered (red), and heterogeneous dense (green) breasts, (i.e., underestimating scattering amplitude and power by 15.1% and 61.1%, 11.1% and 38.9%, 7.9% and 5.6%, respectively), errors in recovered HbT concentration in tumor were 51.4%, 37.8% and 20.2%, respectively. The same trend is observed in Fig. 4(b) when scattered breast density (red) in second simulation case (Fig. 3) was assigned to fatty (blue), homogenous dense (green), and extremely dense (purple) breasts, (i.e., scattering amplitude and power overestimated by -4.5% and -36.4%, 3.6% and 54.5%, 12.5% and 63.6%), respectively, errors in recovered HbT concentration in tumor were 7.3%, 12.6%, and 21.6%, respectively. Also from Fig. 4(b), when optical scattering parameters of scattered breast density were assigned to an extremely dense breast, greater HbT error occurred. We conclude that errors become smaller when the assigned density was closer to the real density in both simulation cases, indicating that assignment of the correct breast density based on MRI is important for maintaining diagnostic accuracy when using only CW data.
Fig. 4. Recovered HbT concentration in different regions and HbT contrast between the tumor and surrounding normal tissue when breast density was mis-assigned in simulation experiments. (a) and (c) The true density is extremely dense in simulation case 1; (b) and (d) The true density is scattered in simulation case 2.
Figure 4(c) and (d) show HbT contrast between tumor and surrounding normal tissue. Table 4 reports HbT contrast between tumor and fibroglandular tissues. Figure 4(c) and (d) and Table 4 indicate mis-assigned scattering properties lead to variations in contrast between tumor to surrounding normal or fibroglandular tissue. Therefore, assigning predetermined scattering properties according to breast density is important.
Table 4. HbT contrast between tumor and fibroglandular tissue for simulation experiments.
3.2 Patient data
Patient Case 1: This example illustrates results obtained from a 61-year-old woman with a 20 mm × 27 mm × 33 mm ductal carcinoma in situ and invasive ductal carcinoma in her right breast. DCE MR images are shown in Fig. 5(b). The breast density was fatty. Her BIRADS score was 5. Average SA and SP values estimated from FD/CW data for the breast were 1.51 and 0.22, while SA and SP values used in CW data reconstruction were 1.17 and 0.14, respectively. HbT images reconstructed from FD/CW and CW only data are shown in Fig. 5(c) and 5(d), respectively. HbT contrasts (shown in Table 5) between tumor and surrounding normal tissue was 1.89, and 3.22 for FD/CW and CW methods, respectively. High values indicate the abnormality was malignant which was confirmed later by pathology.
Fig. 5. Images from patient Case 1: malignancy in fatty breast density. (a) MRI T1; (b) MRI DCE; (c) and (d) are HbT images reconstructed from FD/CW and CW data, respectively. NIRST images are overlaid on the T1 MRI cross-section. The blue lines denote tumor location in (b).
Table 5. HbT contrast estimated from images reconstructed with FD/CW or CW only data.
Patient Case 2: This example presents image data from a 28-year-old woman with a 29 mm × 33 mm × 58 mm fibroadenoma in her right breast. DCE MR images are shown in Fig. 6(b). The breast density was heterogeneous. Her BIRADS score was 3. Average SA and SP values estimated from FD/CW data for the breast were 1.42 and 0.49, while those used in CW data reconstruction were 1.16 and 0.34, respectively. Reconstructed HbT images using FD/CW and CW only data are illustrated in Fig. 6(c) and 6(d), respectively. No significant contrast was observed in HbT at the location of presumed tumor (both values were 1.01, see Table 5), indicating the abnormality was benign which was confirmed later by pathology.
Fig. 6. Images from patient Case 2: a benign abnormality in heterogeneous breast density. (a) MRI T1; (b) MRI DCE; (c) and (d) are HbT images reconstructed from FD/CW data and CW data only, respectively. NIRST images are overlaid on the T1 MRI cross-section. The blue lines denote tumor location in (b).
HbT values in surrounding normal breast were about 30 $\mathrm{\mu}$M in the malignant case (Fig. 5) and about 21 $\mathrm{\mu}$M in the benign case (Fig. 6), and fall within inter-subject differences in normal breast tissue [40]. Average SA and SP values estimated from FD/CW data acquired in these two patient cases are compiled in Table 5. The assigned SA and SP are more than 18% and 30% lower than these quantities.
3.3 Statistical analysis results
When using only CW data, mean HbT contrast in malignant and benign groups was 1.43 and 1.0 with standard deviations 0.55 and 0.05, respectively. They decreased to 1.34 for malignant and remained near 1.0 for benign cases with standard deviations 0.61 and 0.04, respectively, when FD/CW data were applied. Figure 7(a) contains boxplots of HbT contrast reconstructed based on FD/CW or CW only data. Patients were classified into two groups: malignant and benign findings according to pathological results. Significant differences were found in mean HbT contrast (P<0.05) between pathologically-confirmed malignant and benign abnormalities, using either FD/CW or CW data. In CW data case, sensitivity, specificity and diagnostic accuracy were all the same at 87.5%, which was the same as results obtained when using both FD and CW data. As shown in Fig. 7(b), area under the curve (AUC) is higher (94% vs 88%) when CW data was applied relative to FD/CW information. Diagnostic performance for the two data formats is compared in Table 6. In addition, the diagnostic performance of DCE MRI is included in Table 6 and Fig. 7. The data in Table 6, indicate specificity of DCE MRI alone can be improved when MRI-guided NIRST is applied including with only CW measurements. Furthermore, we compared the difference between MRI-guided CW NIRST and DCE MRI ROC curves using the method of DeLong et al. [41], and found it was 0.13, which was not statistically significant (p-value = 0.20).
Fig. 7. Statistical results. (a) Boxplots of HbT contrast in malignant (n=16) and benign (n=8) groups obtained from the images reconstructed from FD/CW and CW data, respectively; (b) ROC curves of DCE MRI and HbT contrast estimated from the images reconstructed from FD/CW and CW data, respectively.
Table 6. Diagnostic Performance for Classifying Breast Lesions Based on Using DCE MRI or FD/CW or CW Data Only Relative to Known Pathology.
4. Discussion and conclusion
Based on both simulated and actual patient image reconstructions, results from this study indicate that HbT contrast can be maintained using only CW data when prior information on scattering properties derived from averages within breast density groups are applied. More specifically, little degradation in HbT contrast was observed in patient data – average HbT contrast in 24 patients was 1.26 with only CW data, which was 3.3% higher than the HbT contrast recovered from FD/CW data. Contrast was variable: 9 patients had more than a 10% decrease whereas the remaining 15 patients had smaller differences. Breasts that exhibited the larger differences in HbT contrast mostly fell in the HD and ED breast density categories.
Theoretically, combined FD/CW data provide more information, and therefore, should achieve higher diagnostic performance than only CW data. However, in a practical imaging session, 12 meter long, 10 mm diameter heavy fiber bundles were used to deliver/collect light signals from the MR console room to the breast in the MRI coil. Although we calibrated phase errors caused by the fiber bundles and fiber bundle holders on the breast very carefully, phase measurements were corrupted by noise arising from optical pathlength changes introduced by the fiber bundle layouts and angles within the holders relative to the breast surface. In our hands, the tradeoff between acquiring richer FD data versus the phase noise incurred, especially at low signal amplitudes, favors CW data which achieved slightly higher (but not statistically different) diagnostic power in the studies reported in this paper (Fig. 7(b)).
Simulation results were generated when scattering properties were fixed as the average of scattering parameters for all breast densities. For simulation Case 1, average HbT values in tumor, fibroglandular, and adipose tissues from CW data were 47.2 ± 9.6 $\mathrm{\mu}$M, 27.5 ± 4.8 $\mathrm{\mu}$M, and 24.2 ± 2.4 $\mathrm{\mu}$M, respectively. HbT contrast between tumor and surrounding normal tissue was 1.88, which is 4.4% higher than that from an extremely dense breast. Thus, HbT errors from average values for all breast densities are larger than those from an extremely dense breast.
Similar results were found in simulation Case 2. Average HbT values reconstructed in tumor, fibroglandular, and adipose tissues from CW data were 29.5 ± 10.9 $\mathrm{\mu}$M, 11.0 ± 5.8 $\mathrm{\mu}$M, and 7.6 ± 3 $\mathrm{\mu}$M, respectively, when average scattering parameter values for all breast densities were used. HbT contrast between tumor and surrounding normal tissue was 3.52. Higher HbT errors and lower contrast were obtained compared to the results reconstructed when scattering parameters were assigned according to breast density.
In our previous studies, slope-dependent calibration has been used [35,36], in which the slope of log of intensity times the source-detector distance determines the absorption coefficient that is fitted at multiple wavelengths to obtain initial estimates of chromophore concentrations. Figure 8 compares estimated HbT concentrations using the direct calibration applied in this study to those obtained with slope-dependent calibration [35,36]. Although direct calibration result in a reduction in mean HbT contrast in malignant and benign cases (from 3.1 & 1.1 with slope dependent calibration to 1.4 & 1.0 with direct calibration), standard deviations also decreased from 2.9 & 0.2 to 0.5 & 0.05, respectively. Sensitivity, specificity and diagnostic accuracy remained at 87.5% independently of calibration method, however, area under the curve (AUC) was slightly higher with direct calibration (0.94 vs 0.9 with slope-dependent calibration), as shown in Fig. 8(b). The significant reduction in standard deviation of HbT contrast in the malignant group, and the slightly higher AUC, suggesting that better stability is obtained by using direct calibration relative to the slope-dependent calibration used previously.
Fig. 8. Statistical results with different data calibration algorithms. (a) Boxplots of HbT contrast in malignant (n=16) and benign (n=8) groups obtained from reconstructed images with direct (Direct) and slope-dependent (Slope) calibration methods, respectively; (b) ROC curves of DCE MRI and HbT contrast estimated from the images reconstructed from direct and slope-dependent calibration methods, respectively.
Although the small sample size of this study may cause some limitations of the statistical analysis, MRI alone tends to have a higher false positive rate for breast cancer diagnosis. As shown in Table 6, DCE MRI alone had an AUC of 0.81 with specificity and sensitivity of 62.5% and 100%, respectively; while DCE MRI-guided CW NIRST alone had an AUC of 0.94 with specificity and sensitivity of 87.5% and 87.5%, respectively. There is no significant difference between DCE MRI-guided CW NIRST versus DCE MRI performance in terms of AUC. However, by adding optical data to MRI, the false positive rate of DCE MRI could be reduced. When DCE MRI-guided NIRST based on CW data and DCE MRI were both used for diagnostic power analysis, the specificity and AUC can be improved to 100% and 0.97, while maintaining the sensitivity of 87.5%.
In conclusion, results from this study show that MRI guided NIRST based on CW data and scattering properties assigned to average values from breast density categories as prior information, can differentiate benign and malignant legions with diagnostic power similar to images reconstructed from full FD/CW data. The value of the CW MR-guided NIRST approach is its simplicity and low cost relative to acquiring frequency domain phase measurements.
While recovering scattering properties for different tissue types may yield more accurate images, image reconstruction, in this case, requires segmentation in order to assign different scattering parameters to different types of tissue. The segmentation step adds processing complexity and reduces objectivity when combining the image information. Additionally, segmentation can be time-consuming and requires experience to avoid introduction of bias or error. Thus, a direct regularization imaging method was utilized in this study to avoid the need to perform specific tissue-type segmentation.
Moreover, our results reveal that assigning the same scattering parameters to fibroglandular and adipose tissues yields satisfactory results, and simplifies the computational pipeline significantly. Segmenting fibroglandular tissue from MRI T1 images would provide an accurate breast density percentage obtainable through fast and automated algorithms. Future work will evaluate the effects of using MRI-derived fat-fibroglandular percentages to estimate breast density from which breast scattering properties are determined on optically differentiating malignant from benign tumors.
Funding
National Natural Science Foundation of China (81871394, 82171992); National Cancer Institute (R01 CA069544, R01 CA176086).
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. N. Shah, A. Cerussi, C. Eker, J. Espinoza, J. Butler, J. Fishkin, R. Hornung, and B. Tromberg, “Noninvasive functional optical spectroscopy of human breast tissue,” Proc. Natl. Acad. Sci. 98(8), 4420–4425 (2001). [CrossRef]
2. V. Ntziachristos, A. G. Yodh, M. Schnall, and B. Chance, “Concurrent MRI and diffuse optical tomography of breast after indocyanine green enhancement,” Proc. Natl. Acad. Sci. 97(6), 2767–2772 (2000). [CrossRef]
3. B. W. Pogue, S. P. Poplack, T. O. McBride, W. A. Wells, K. S. Osterman, U. L. Osterberg, and K. D. Paulsen, “Quantitative hemoglobin tomography with diffuse near-infrared spectroscopy: pilot results in the breast,” Radiology 218(1), 261–266 (2001). [CrossRef]
4. A. K. E. Devolli-Disha, S. Manxhuka-Kerliu, and H. Ymeri, “Comparative accuracy of mammography and ultrasound in women with breast symptoms according to age and breast density,” Bosn J of Basic Med Sci 9(2), 131–136 (2009). [CrossRef]
5. Q. Fang, J. Selb, S. A. Carp, G. Boverman, E. L. Miller, D. H. Brooks, R. H. Moore, D. B. Kopans, and D. A. Boas, “Combined optical and X-ray tomosynthesis breast imaging,” Radiology 258(1), 89–97 (2011). [CrossRef]
6. Q. Zhu, S. Tannenbaum, P. Hegde, M. Kane, C. Xu, and S. H. Kurtzman, “Noninvasive monitoring of breast cancer during neoadjuvant chemotherapy using optical tomography with ultrasound localization,” Neoplasia 10(10), 1028–1040 (2008). [CrossRef]
7. S. Jiang, B. W. Pogue, T. O. McBride, and K. D. Paulsen, “Quantitative analysis of near-infrared tomography: sensitivity to the tissue-simulating precalibration phantom,” J. Biomed. Opt. 8(2), 308–315 (2003). [CrossRef]
8. C. M. Carpenter, S. Srinivasan, B. W. Pogue, and K. D. Paulsen, “Methodology development for three-dimensional MR-guided near infrared spectroscopy of breast tumors,” Opt. Express 16(22), 17903–17914 (2008). [CrossRef]
9. M. A. Mastanduno, J. Xu, F. El-Ghussein, S. Jiang, H. Yin, Y. Zhao, K. Wang, F. Ren, J. Gui, B. W. Pogue, and K. D. Paulsen, “MR-guided near-infrared spectral tomography increases diagnostic performance of breast MRI,” Clin. Cancer Res. 21(17), 3906–3912 (2015). [CrossRef]
10. J. Feng, J. Xu, S. Jiang, H. Yin, Y. Zhao, J. Gui, K. Wang, X. Lv, F. Ren, B. W. Pogue, and K. D. Paulsen, “Addition of T2-guided optical tomography improves noncontrast breast magnetic resonance imaging diagnosis,” Breast Cancer Res. 19(1), 117 (2017). [CrossRef]
11. L. Zhang, Y. Zhao, S. Jiang, B. W. Pogue, and K. D. Paulsen, “Direct regularization from co-registered anatomical images for MRI-guided near-infrared spectral tomographic image reconstruction,” Biomed. Opt. Express 6(9), 3618–3630 (2015). [CrossRef]
12. V. Krishnaswamy, K. E. Michaelsen, B. W. Pogue, S. P. Poplack, I. Shaw, K. Defrietas, K. Brooks, and K. D. Paulsen, “A digital x-ray tomosynthesis coupled near infrared spectral tomography system for dual-modality breast imaging,” Opt. Express 20(17), 19125–19136 (2012). [CrossRef]
13. J. Feng, S. Jiang, J. Xu, Y. Zhao, B. W. Pogue, and K. D. Paulsen, “Multiobjective guided priors improve the accuracy of near-infrared spectral tomography for breast imaging,” J. Biomed. Opt. 21(9), 090506 (2016). [CrossRef]
14. Y. Zhao, M. A. Mastanduno, S. Jiang, F. EI-Ghussein, J. Gui, B. W. Pogue, and K. D. Paulsen, “Optimization of image reconstruction for magnetic resonance imaging–guided near-infrared diffuse optical spectroscopy in breast,” J. Biomed. Opt. 20(5), 056009 (2015). [CrossRef]
15. Y. Zhao, B. W. Pogue, S. J. Haider, J. Gui, R. M. diFlorio-Alexander, K. D. Paulsen, and S. Jiang, “Portable, parallel 9-wavelength near-infrared spectral tomography (NIRST) system for efficient characterization of breast cancer within the clinical oncology infusion suite,” Biomed. Opt. Express 7(6), 2186–2201 (2016). [CrossRef]
16. S. Jiang, B. W. Pogue, P. A. Kaufman, J. Gui, M. Jermyn, T. E. Frazee, S. P. Poplack, R. DiFlorio-Alexander, W. A. Wells, and K. D. Paulsen, “Predicting breast tumor response to neoadjuvant chemotherapy with diffuse optical spectroscopic tomography prior to treatment,” Clin. Cancer Res. 20(23), 6006–6015 (2014). [CrossRef]
17. S. R. Arridge, M. Cope, and D. T. Delpy, “The theoretical basis for the determination of optical pathlengths in tissue: temporal and frequency analysis,” Phys. Med. Biol. 37(7), 1531–1560 (1992). [CrossRef]
18. J. Wang, B. W. Pogue, S. Jiang, and K. D. Paulsen, “Near-infrared tomography of breast cancer hemoglobin, water, lipid, and scattering using combined frequency domain and CW measurement,” Opt. Lett. 35(1), 82–84 (2010). [CrossRef]
19. J. Wang, S. Jiang, Z. Li, R. M. diFlorio-Alexander, R. J. Barth, P. A. Kaufman, B. W. Pogue, and K. D. Paulsen, “In vivo quantitative imaging of normal and cancerous breast tissue using broadband diffuse optical tomography,” Med. Phys. 37(7Part1), 3715–3724 (2010). [CrossRef]
20. R. M. P. Doornbos, R. Lang, M. C. Aalders, F. W. Cross, and H. J. C. M. Sterenborg, “The determination of in vivo human tissue optical properties and absolute chromophore concentrations using spatially resolved steady-state diffuse reflectance spectroscopy,” Phys. Med. Biol. 44(4), 967–981 (1999). [CrossRef]
21. J. B. Fishkin, O. Coquoz, E. R. Anderson, M. Brenner, and B. J. Tromberg, “Frequency-domain photon migration measurements of normal and malignant tissue optical properties in a human subject,” Appl. Opt. 36(1), 10–20 (1997). [CrossRef]
22. C. D’Andrea, L. Spinelli, A. Bassi, A. Giusto, D. Contini, J. Swartling, A. Torricelli, and R. Cubeddu, “Time-resolved spectrally constrained method for the quantification of chromophore concentrations and scattering parameters in diffusing media,” Opt. Express 14(5), 1888–1898 (2006). [CrossRef]
23. S. Srinivasan, B. W. Pogue, S. Jiang, H. Dehghani, C. Kogel, S. Soho, J. J. Gibson, T. D. Tosteson, S. P. Poplack, and K. D. Paulsen, “In vivo hemoglobin and water concentrations, oxygen saturation, and scattering estimates from near-infrared breast tomography using spectral reconstruction,” Academic Radiology 44(4), 967–981 (1999). [CrossRef]
24. M. A. Mastanduno, J. Xu, F. El-Ghussein, S. Jiang, H. Yin, Y. Zhao, K. E. Michaelsen, K. Wang, F. Ren, B. W. Pogue, and K. D. Paulsen, “Sensitivity of MRI-guided near-infrared spectroscopy clinical breast exam data and its impact on diagnostic performance,” Biomed. Opt. Express 5(9), 3103–3115 (2014). [CrossRef]
25. K. Michaelsen, V. Krishnaswamy, B. W. Pogue, S. P. Poplack, and K. D. Paulsen, “Near-infrared spectral tomography integrated with digital breast tomosynthesis: effects of tissue scattering on optical data acquisition design,” Med. Phys. 39(7Part2), 4579–4587 (2012). [CrossRef]
26. D. J. Davies, M. Clancy, D. Lighter, G. M. Balanos, S. J. E. Lucas, H. Dehghani, Z. Su, M. Forcione, and A. Belli, “Frequency-domain vs continuous-wave near-infrared spectroscopy devices: a comparison of clinically viable monitors in controlled hypoxia,” J Clin Monit Comput 31(5), 967–974 (2017). [CrossRef]
27. J. P. Culver, R. Choe, M. J. Holboke, L. Zubkov, T. Durduran, A. Slemp, V. Ntziachristos, B. Chance, and A. G. Yodh, “Three-dimensional diffuse optical tomography in the parallel plane transmission geometry: evaluation of a hybrid frequency domain/continuous wave clinical system for breast imaging,” Med. Phys. 31(5), 967–974 (2017). [CrossRef]
28. S. Niu, Q. Zhu, Y. Jiang, J. Zhu, M. Xiao, S. You, W. Zhou, and Y. Xiao, “Correlations among ultrasound-guided diffuse optical tomography, microvessel density, and breast cancer prognosis,” J Ultrasound Med 37(4), 833–842 (2018). [CrossRef]
29. D. R. Leff, O. J. Warren, L. C. Enfield, A. Gibson, T. Athanasiou, D. K. Patten, J. Hebden, G. Yang, and A. Darzi, “Diffuse optical imaging of the healthy and diseased breast: a systematic review,” Breast Cancer Res. Treat. 108(1), 9–22 (2008). [CrossRef]
30. E. A. Lim, J. E. Gunther, H. K. Kim, M. Flexman, H. Hibshoosh, K. Crew, B. Taback, J. Campbell, K. Kalinsky, A. Hielscher, and D. L. Hershman, “Diffuse optical tomography changes correlate with residual cancer burden after neoadjuvant chemotherapy in breast cancer patients,” Breast Cancer Res. Treat. 162(3), 533–540 (2017). [CrossRef]
31. L. Spinelli, A. Torricelli, A. Pifferi, P. Taroni, G. Danesini, and R. Cubeddu, “Characterization of female breast lesions from multi-wavelength time-resolved optical mammography,” Phys. Med. Biol. 50(11), 2489–2502 (2005). [CrossRef]
32. A. Cerussi, N. Shah, D. Hsiang, A. Durkin, J. Butler, and B. J. Tromberg, “In vivo absorption, scattering, and physiologic properties of 58 malignant breast tumors determined by broadband diffuse optical spectroscopy,” J. Biomed. Opt. 11(4), 044005 (2006). [CrossRef]
33. S. Srinivasan, B. W. Pogue, S. Jiang, H. Dehghani, C. Kogel, S. Soho, J. J. Gibson, T. D. Tosteson, S. P. Poplack, and K. D. Paulsen, “Interpreting hemoglobin and water concentration, oxy-gen saturation, and scattering measured in vivo by near-infrared breast tomography,” Proc. Natl. Acad. Sci. 100(21), 12349–12354 (2003). [CrossRef]
34. H. Dehghani, M. E. Eames, P. K. Yalavarthy, S. C. Davis, S. Srinivasan, C. M. Carpenter, B. W. Pogue, and K. D. Paulsen, “Near infrared optical tomography using NIRFAST: Algorithm for numerical model and image reconstruction,” Commun. Numer. Meth. Engng. 25(6), 711–732 (2009). [CrossRef]
35. B. W. Pogue, K. D. Paulsen, C. Abele, and H. Kaufman, “Calibration of near-infrared frequency-domain tissue spectroscopy for absolute absorption coefficient quantitation in neonatal head-simulating phantoms,” J. Biomed. Opt. 5(2), 185–193 (2000). [CrossRef]
36. S. Jiang, B. W. Pogue, T. O. Mcbride, M. M. Doyley, S. P. Poplack, and K. D. Paulsen, “Near-infrared breast tomography calibration with optoelastic tissue simulating phantoms,” J. Electron. Imaging 12(4), 613–620 (2003). [CrossRef]
37. V. Ntziachristos, A. G. Yodh, M. D. Schnall, and B. Chance, “MRI-Guided diffuse optical spectroscopy of malignant and benign breast lesions,” Neoplasia 4(4), 347–354 (2002). [CrossRef]
38. P. K. Yalavarthy, B. W. Pogue, H. Dehghani, C. M. Carpenter, S. Jiang, and K. D. Paulsen, “Structural information within regularization matrices improves near infrared diffuse optical tomography,” Opt. Express 4(4), 347–354 (2002). [CrossRef]
39. J. Feng, Sh. Jiang, B. Pogue, and P. Keith, “Weighting function effects in a direct regularization method for image-guided near- infrared spectral tomography of breast cancer,” Biomed. Opt. Express 9(7), 3266–3283 (2018). [CrossRef]
40. N. Shah, A. E. Cerussi, D. Jakubowski, D. Hsiang, J. Butler, and B. J. Tromberg, “Spatial variations in optical and physiological properties of healthy breast tissue,” J. Biomed. Opt. 9(3), 534–540 (2004). [CrossRef]
41. E. R. DeLong, D. M. DeLong, and D. L. Clarke-Pearson, “Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach,” Biometrics 44(3), 837–845 (1988). [CrossRef]
