Proceedings of the 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference Shanghai, China, September 1-4, 2005
Localization and Functional Parameter Reconstruction of Suspicious Breast Lesions by Near Infrared/Ultrasound Dual Modal Imaging Ronald X. Xu*, Bo Qiang
John O. Olsen, Stephen P. Povoski, Lisa D. Yee
Jimmy Mao
Biomedical Engineering Center The Ohio State University Columbus, OH 43210
James Cancer Hospital and Solove Research Institute The Ohio State University Columbus, OH 43210
ViOptix, Inc., Fremont, CA 94538
Abstract - A novel imaging scheme integrating a continuous wave (CW) hand held near infrared tissue imager and a portable ultrasound probe was proposed for evaluation of suspicious breast lesions. A new methodology was developed to reconstruct functional properties of the lesion and the surrounding tissue based on the optical measurement of the diffusive light and the ultrasound measurement of the tumor morphology. The first order Born approximation was used to solve the absorption coefficients for both the tumor tissue and the background tissue assuming that optical properties within each tissue type are uniform. A compression force was applied by the handheld near infrared imager to the breast tissue in order to enhance the optical contrast. The force was monitored by a built-in load cell. 2D projection algorithm was used to find the center of the tumor and to co-register the near infrared image with the ultrasound image. The proposed imaging scheme was tested by a clinical trial where both near infrared and ultrasound data were collected on subjects’ breast tissues with embedded suspicious lesions. 3D image algorithm on single patient successfully reconstructed the oxygen saturation and the hemoglobin concentration in both the tumor area and the surrounding tissue. More data collection and analysis are required to confirm this imaging scheme*.
I. INTRODUCTION Many cancers alter local vascularization and metabolism, producing early physiological changes in the level of blood circulation and oxygenation over an area of surrounding tissue. For example, hemoglobin de-saturation in malignant tumors may be increased due to the high oxygen demand and blood volume may be increased over that of normal background tissue due to the greater vascularization and metabolic needs [1-3]. Near infrared diffuse optical imaging and spectroscopy (DOIS) can quantify these physiological parameters and reconstruct the tissue heterogeneity by noninvasive optical measurement. Substantially greater hemoglobin concentration has been reported for breast tissues with cancerous lesions. Various criteria have been defined in order to differentiate cancerous and normal breast tissues based on their oxygen saturation and hemoglobin
concentration contrasts[4]. Despite the fact that DOIS is non-invasive, non-radioactive, and able to measure tissue functional properties in real time, it has not been widely adopted clinically. This is partially due to the following reasons: 1) the imaging system is complicated and sensitive to tissue boundary condition; and 2) technology is limited in depth and spatial resolution, partially because of the exponential attenuation of light in biological tissue and partially because of the heavily under-determined diffusion equations. In the recent past, we have developed handheld a continuous wave (CW) imaging system called P-Scan tissue imager for breast cancer diagnosis [5-7]. It consists of a handheld probe about 5.5 cmX5.5 cmX10.2 cm with embedded laser diode and photo diode modules, drivers, microprocessor, and an embedded personal computer (PC) with a liquid crystal display (LCD) monitor, as shown in Figure 1. The imager is also equipped with a compression sensor in order to monitor the tissue dynamic response to external pressure stimuli. The validity of the P-Scan tissue imager was tested in a pilot clinical trial of 50 subjects [7]. The test results showed improved sensitivity and specificity in conjunction with standard mammography.
Figure 1 Left: P-Scan tissue imaging system, Right: P-Scan probe
*
Corresponding author: Ronald Xu, Biomedical Engineering center, 270 Bevis Hall, Columbus, OH 43210, USA. Tel: +1 614 688 3635; Fax: +1 614 292 7301; Email:
[email protected]
0-7803-8740-6/05/$20.00 ©2005 IEEE.
4473
Figure 2 Terason 2000 ultrasound system used for clinical study.
This paper proposes a Near Infrared/ Ultrasound dual imaging modality that combines a P-Scan imager shown in Figure 1 and a Terason 2000 ultrasound system shown in Figure 2. The ultrasound probe images the cross section of the breast tissue heterogeneity that provides the localization and morphological guidance to the tumor optical property reconstruction. The methods of introducing other imaging modalities such as MRI and ultrasound to enhance the accuracy and the spatial resolution of the near infrared modality have been explored extensively by other research groups [8-11]. Our approach is different from theirs in that: 1) a low cost, handheld, CW imaging system is used for near infrared data collection; 2) breast tissue is divided into tumor and background zones and homogeneity is assumed within each zone; 3) near infrared image and ultrasound image are co-registered by a frame marker; 4) Tumor’s lateral center is further refined on a 2D projected near infrared image prior to 3D reconstruction; and 5) compression force at a controlled level is applied in order to enhance the optical contrast between tumor and surrounding tissue. II. ALGORITHM
§ GU (rs1 , rd 1 ) · ¨ ¸ ¨ ¸ ¨ GU (r , r ) ¸ sm dm ¹ m ,1 ©
· ¸ ¸ Eq. 2 ¸ ¸ ¸ ¸ ¸ ¸¸ ¹ n ,1
Where P a 0 is the background absorption, P a T is the tumor absorption, and Wij G (rdi , r j )U 0 (rsi , r j )va 3 / D is the contribution (weight) of the jth voxel of the detected media to the measurement of the ith source-detector pair. This equation group involves only 2 unknowns ( P a 0 , P a T ) and can be solved as a deterministic system for the sourcedetector configuration of a Pscan imager. In order to enhance the optical contrast between tumor and surrounding tissues, a compression force was gradually applied to and then released from the breast tissue following a profile like Figure 4. The pressure value is monitored real time by a built-in pressure sensor. Steps for image reconstruction are demonstrated by a flow chart as sketched below:
The reconstruction algorithm was based on the first order Born approximation with the flowing assumptions: 1) the absorption contrast between tumor tissue zone and surrounding tissue zone is substantially larger than their individual heterogeneity so that homogeneity is assumed within each tissue type; 2) the absorption contrast and the size of the tumor is within the range where the Born approximation is valid; 3) tissue scattering property is preset and the scattering difference between tumor and surrounding tissue is negligibl;. 4) tumor is reconstructed as an ellipsoid with equal short axis lengths based on the ultrasound cross-sectional image; and 5) tumor depth does not experience significant variation during compression. The first order Born approximation of the diffuse equation under small absorption perturbation is expressed as: * * * * * * QGP a 3 Eq. 1 GU (rs , rd ) ³ G (r , rd )U 0 (r , rs ) dr D Where U 0 is the optical measurement on a homogenous background, GU is the measurement perturbation due to embedded absorbing object, G (r*, r*d ) is the Green function for the Radiative Transfer Equation, and GP a P a P a is the absorption contrast of the embedded object. o
§ ¨ ¨ 0 0 § W11 ( P a ) W1n ( P a ) · ¨ GP T ¨ ¸ ¨ a ¨ ¸ ¨ ¨¨Wm1 ( P a 0 ) Wmn ( P a 0 ) ¸¸ ¨0 © ¹ m,n ¨ GP T ¨¨ a ©
b
Reconstruct tumor depth, long axis length, short axis length and tumor orientation from the ultrasound image.
Calculate the 2D projection of tissue oxygen saturation [SO2] and hemoglobin concentration [Hbt] at different pressure levels.
Find an 2D projected image (image A) of [Hbt] at a high pressure level where tumor shows high contrast with the surrounding tissue.
Locate lateral center of the tumor in image A
Inverse reconstruction of tumor and surrounding tissue absorption properties ( P a 0 and P a T ) by iterative first order Born approximation.
The discrete form of the above equation can be expressed as: Calculation of [SO2] and [Hbt] distributions Figure 3 Flow chart for NIR/US dual modal image reconstruction
4474
The above image reconstruction scheme is demonstrated by a clinical example described in the next section.
Compression pressure profile for NIR imaging 1.6
1.4
III. CLINICAL DATA COLLECTION AND ANALYSIS A clinical protocol for evaluation of suspicious breast lesions using near infrared/ultrasound dual modal imaging technique was approved by the Ohio State University Medical Center Cancer Institutional Review Board (protocol No: OSU04117, IRB No: 2005C005). Clinical data collection was performed at JamesCare Breast Care Center in Dublin, Ohio. Figure 4 shows the clinical trial setup where a portable ultrasound probe, a handheld tissue imager and a laptop computer were placed on a mobile cart. Nonpregnant female subjects at age 18 or older, with breast lesions identified by mammography or ultrasound (BIRAD 4 or 5) were recruited for the test. The test was arranged right after clinical ultrasound and before a scheduled biopsy procedure. The subject was in a supine position and an Aluminum frame marker (shown in Figure 4 as item E.) was placed on the top of suspicious lesion for co-registration between ultrasound and near infrared images. At the beginning of the test, an ultrasound image was captured by a Terason 2000 portable probe at the center of the suspicious lesion. Then a P-Scan imager was placed in the Aluminum frame marker for continuous data collection for about 20 seconds while a mild pressure (maximal 1.5Kg) like that in Figure 5 was gradually applied to and released from the breast tissue. The data collected from the near infrared imager and the ultrasound probe were further analyzed for tumor oxygen saturation and hemoglobin concentration contrasts. The results will be compared with other clinical imaging and biopsy results in order to determine the diagnostic criteria.
Pressure level (Kg)
1.2
1
0.8
0.6
0.4
0.2 20
22
24
26
28
30 32 Sample No
34
36
38
40
Figure 5 Compression pressure applied to the tissue
The imaging and data processing procedure is demonstrated here by single patient example. This is a 44 year old woman with extremely dense breasts. A suspicious right breast lesion was detected by mammogram and ultrasound. On ultrasound, this was located in the 9-10 o’clock axis of the right breast in zone 2, measuring 0.58cm x 0.41cm x 0.51cm in size (Figure 6). The lesion size and location was abstracted by a Matlab program and illustrated in Figure 7.
Figure 6 Ultrasound image of a right breast lesion
E. Aluminum frame marker Reconstructed tumor model: X0: 0 cm Y0: 0.07cm Z0: 1.03 cm long axis length a: 0.56cm short axis length b: 0.35cm tumor orientation: 11º Figure 7 Reconstruction of tumor geometry and localization from the ultrasound image
Figure 4 Clinical trial setup
4475
Figure 7 demonstrates that the tumor size and the location can be abstracted from the ultrasound image captured by the portable ultrasound probe. However, the lateral position of the lump may not be accurate for near infrared reconstruction purpose due to the position mismatch between the ultrasound probe and the P-Scan probe. Therefore, we did a 2D projected imaging following a scheme discussed in [12] to refine the tumor lateral location. Figure 8 plots the 2D projected map of [Hbt] in response to the compression pressure profile as shown in Figure 5. 0.227 kg
0.238 kg
0.395 kg
0.552 kg
0.608 kg
0.619 kg
0.753 kg
0.865 kg
0.887 kg
0.954 kg
1.1 kg
1.111 kg
1.256 kg
1.2 kg
0.775 kg
0.787 kg
0.608 kg
0.697 kg
0.552 kg
0.406 kg
functional parameters for a suspicious breast lesion and surrounding normal breast tissue. Data analysis on more patients is required in order to validate this imaging scheme and to develop proper criteria that may differentiate normal and abnormal breast lesions. In addition, current work does not address the tumor depth change due to compression, which will be considered in the future analysis. Further directions include the development of more efficient algorithm for 3D inverse reconstruction, data analysis for the [StO2]/[Hbt] dynamics in response to external pressure stimuli as additional diagnostic criteria [6], and the development of the hybrid imaging system with more wavelengths and higher resolution. V. ACKKNOWLEDGEMENT This research is supported in part by ViOptix, Inc. under contract No. GRT00001204. Authors appreciate biostatistical support from Dr. Donn Young, and clinical support from the following staff members at JamesCare in Dublin: Jo Clarke, Joanne Lester, Janell Tucker, Maria Sanchez, Lori Creiglow, and Laura Kirk-Fetsko. VI. REFERENCES 1.
Figure 8 2D projection of [Hbt] response to external compressive pressure 2.
It is evident from Figure 8 that blood was pushed away from the tissue and the tumor contrast reaches its maximum at high pressure levels. We choose the optical measurement at the peak compression pressure (1.25Kg) for lateral tumor localization and 3D image reconstruction. The 3D image reconstruction follows the scheme discussed in section II. The weight matrix in Eq. 2 was sparsed first in order to simplify computation. Constant scattering coefficients of P s '690 = 10cm-1 and P s '830 = 7.8 cm-1 were used for both tumor and surrounding tissues. An objective function was defined to minimize the difference between the measured and the simulated optical densities. The iterative optimization resulted in a converged solution of P a 0 _ 690 = 0.0865cm-1, Pa _ 830 =0.2176cm-1, Pa _ 690 = 0.1100cm1 and PaT _ 830 = 0.2450cm-1(830cm-1). These absorption 0
3. 4. 5. 6. 7. 8.
T
coefficients correspond to an oxygen saturation level of 75.48% and 70.91%, and a hemoglobin concentration level of 5.3PM and 1.19PM for surrounding tissue and tumor respectively.
9. 10.
11.
IV. CONCLUSION AND FUTURE WORK We have presented a novel imaging scheme for evaluation of breast suspicious lesions by combining near infrared and ultrasound modalities. We have demonstrated this scheme through a clinical example where both ultrasound data and near infrared data were collected for 3D reconstruction of
12.
4476
Folkman, J., What is the evidence that tumors are angiogenesis dependent. J. Natl. Cancer Inst., 1990. 82(1): p. 4-6. Vaupel, P., A. Mayer, et al., Oxygenation gain factor: a novel parameter characterizing the association between hemoglobin level and the oxygenation status of breast cancers. Cancer Res, 2003. 63(22): p. 7634-7. Zhou, S., Y. Chen, et al., A portable dual wavelength amplitude cancellation image system for the determination of human breast tumor. Proc. SPIE, 1999. 3597: p. 571-579. Tromberg, B.J., N. Shah, et al., Non-invasive in vivo characterization of breast tumors using photon migration spectroscopy. Neoplasia, 2000. 2(1-2): p. 26-40. Cheng, X., X. Xu, et al., A novel optical scanning system for breast cancer imaging. Proc. SPIE, 2001. 4244: p. 468-473. Cheng, X. and X. Xu, Study of the pressure effect in near infrared spectrscopy. Proc. SPIE, 2003. 4955: p. 397-406. Cheng, X., J.-m. Mao, et al., Breast Cancer Detection by Mapping Hemoglobin Concentration and Oxygen Saturation. Applied Optics, 2003. 42(31): p. 6412-6421. Ntziachristos, V., A.G. Yodh, et al., Concurrent MRI and diffuse optical tomography of breast after indocyanine green enhancement. Proc Natl Acad Sci U S A, 2000. 97(6): p. 276772. Holboke, M.J., B.J. Tromberg, et al., Three-dimensional diffuse optical mammography with ultrasound localization in a human subject. J Biomed Opt, 2000. 5(2): p. 237-47. Zhu, Q., M. Huang, et al., Ultrasound-guided optical tomographic imaging of malignant and benign breast lesions: initial clinical results of 19 cases. Neoplasia, 2003. 5(5): p. 37988. Zhu, Q., E. Conant, et al., Optical imaging as an adjunct to sonograph in differentiating benign from malignant breast lesions. J Biomed Opt, 2000. 5(2): p. 229-36. Ronald Xu, J.M., Near Infrared Imaging of Tissue Heterogeneity: Probe Design and Sensitivity Analysis. IEEE EMBS Conference, 2005, in print.
