Published online before print September 21, 2007, 10.1148/radiol.2452060906
(Radiology 2007;245:556-566.)
© RSNA, 2007
Breast Cancers: Noninvasive Method of Preoperative Localization with Three-dimensional US and Surface Contour Mapping1
Charlotte J. C. Cash, MRCP, FRCR,
Charlotte E. Coles, MRCP, FRCR,
Graham M. Treece, MA, PhD,
Anand D. Purushotham, MD, FRCS,
Peter Britton, FRCS, FRCR,
Ruchi Sinnatamby, MRCP, FRCR,
Andrew H. Gee, MA, PhD, and
Richard W. Prager, MA, PhD
1 From the Departments of Radiology (C.J.C.C.), Oncology (C.E.C.), Surgery (A.D.P.), and Engineering (G.M.T., A.H.G., R.W.P.), Cambridge University, Hills Road, Cambridge CB22QQ, England; and Department of Radiology, Addenbrooke's Hospital NHS Trust, Cambridge, England (P.B., R.S.). Received May 24, 2006; revision requested July 27; revision received October 20; accepted November 21; final version accepted April 2, 2007.
Address correspondence to C.J.C.C. (e-mail: cjcc2{at}ukonline.co.uk).
 |
ABSTRACT
|
|---|
Formal ethical approval was granted by the local research ethics committee; all participants gave written consent. The purpose of the study was to prospectively evaluate the feasibility of a noninvasive method of breast tumor localization in 25 participants, based on the coregistration of three-dimensional (3D) ultrasonographic (US) data with surface contour data obtained by using a 3D laser camera. The tumor is segmented from the US data, and a surface-rendered 3D image of the tumor, in relation to the breast surface contour, is produced. From a personal computer in the operating room, the surgeon can dynamically view a 3D image of the tumor within the breast. This noninvasive method was equivalent to conventional techniques in 18 of 25 patients but was less successful in larger-breasted patients. In selected patients, this localization method could provide an alternative to conventional invasive techniques and can offer both spatial localization and tumor morphology.
© RSNA, 2007
|
INTRODUCTION
|
|---|
Impalpable breast cancers frequently are suitable for breast-conserving surgery. The ideal localization procedure should accurately guide the surgeon to the lesion, result in the removal of adequate breast tissue with satisfactory excision margins, and be well tolerated by the patient. Numerous techniques have been employed to localize impalpable lesions prior to surgical excision.
Hook-wire localization is a widely used technique (1,2) and involves placement of a wire through or adjacent to the tumor with mammographic or ultrasonographic (US) guidance. There are a number of disadvantages to the patient, which include discomfort and vasovagal syncope (3–5). Technical and clinical complications, which include wire migration (5,6), pneumothorax (7), and dislodgement or transection (8), have been described. Also, the skin entry site of the wire is determined by the radiologist, and this site may not be the most optimal site for the surgeon.
A successful target rate of greater than 95% has been reported to be achievable with the hook wire (9); however, "success" may vary from institution to institution depending on local policy in regard to the accepted minimum clear margin. Many of the disadvantages of the hook wire are eliminated with radioactive occult lesion localization (ROLL) (10), and lower reexcision rates have been reported with ROLL compared with the rates with hook-wire localization (11). With ROLL, however, the procedure remains invasive, and there is the potential for radiation exposure to the patient and surgeon. Although intraoperative US is reported to be as accurate as (12) or even more accurate than hook-wire localization (13), the technique relies on the surgeon's training in US, or a sonographer needs to be employed in the operating room, which may not be cost-effective. If lesions are difficult to detect, there may be inevitable delays in the surgical schedule.
The purpose of our study was to prospectively evaluate the feasibility of a noninvasive method of breast tumor localization, which is based on the coregistration of three-dimensional (3D) US data with surface contour data obtained by using a 3D laser camera.
|
MATERIALS AND METHODS
|
|---|
Three-dimensional US System
Details of the system.—The 3D US images used in this study were produced by using a freehand high-spatial-resolution 3D US system (14). The gray-scale images were produced by using a US machine (Aplio 80; Toshiba Medical Systems, Tustin, Calif) with a 12-MHz matrix transducer. The images produced had high spatial resolution with a pixel size of approximately 0.1 x 0.1 mm and were acquired with a 4-cm depth setting. With the freehand 3D US system, an optical position sensor (Polaris; Northern Digital, Waterloo, Ontario, Canada) was used, and this sensor was used to continually track the orientation of 15 infrared-emitting diodes attached to the transducer. US images, together with their spatial orientation data, were simultaneously recorded to an 800-MHz personal computer (World of Computers, Cambridge, England) equipped with research software (Stradx; Department of Engineering, Cambridge University, Cambridge, England [http://mi.eng.cam.ac.uk/
rwp/stradx]). This noncommercial research software was also used to analyze and display the 3D data and has been described in detail elsewhere (14).
Coregistration: alignment of 3D US data with 3D laser surface contour data.—A tracked 3D pointer was developed that could record the location of points in space in the same coordinate system as the 3D US data by using the same position-sensing technology used to locate the US transducer. Three-dimensional surface contour data were acquired with a tripod-mounted laser camera (Vivid 700; Konica Minolta 3D, Milton Keynes, England) positioned approximately 1.5 m away from the object of interest by an investigator (C.E.C., with 2 years of laser camera experience). The camera has a resolution of approximately 1 mm and a scanning time of 0.6 second. In addition to the acquisition of surface data, this system simultaneously takes a color photograph of the area that is registered to the surface data. Therefore, fiducial marks (crosses) placed on a surface can be recorded with the 3D US system by using the 3D pointer and also with the 3D laser system derived visually by using the color photograph. The process of registration involves matching the two sets of data from the two different coordinate systems (ie, points recorded by the 3D US system are coregistered with the crosses on the photograph produced with the laser camera). The 3D US data and the laser surface contour data are then aligned by iterative minimization of the difference between the locations of points recorded by using the tracked 3D pointer with the same points seen as crosses on the photograph, and this process results in a rigid point-based registration. The registration procedure is performed by using the research software mentioned previously.
Validation Work
Extensive validation work to assess the spatial accuracy of the 3D US system alone has previously been performed. Results from this work estimate its point location accuracy to be within 0.5 mm (14). For the purpose of this study, the coregistration process was assessed by identifying the optimum number and color of fiducial crosses that produce the most accurate coregistration.
Color of coregistration (fiducial) mark.—The laser camera generates the 3D surface contour data by sweeping a laser over a surface and receiving its reflection. Abrupt changes in the color of the surface (eg, fiducial crosses made on the patient's skin) result in a variation in the reflected light and, therefore, affect the accuracy of the reconstructed surface. Since marks on the skin were a necessary part of the registration process, the accuracy of the reconstructed surface at these points was critical. Two investigators (G.M.T., C.E.C.) performed a series of validation experiments to compare the same skin surface with and without different colored crosses (blue, green, pink, and red) and to identify which color causes the least change to the reconstructed laser surface. An objective measurement of the effect of color on the reconstructed surface was made by measuring 100 equivalent loci between the two reconstructed laser surfaces, thus providing a range of distances for each color. The reconstruction accuracy itself was assessed by comparing the reconstruction of two identical surfaces with no marks.
Number of coregistration (fiducial) marks.—The coregistration accuracy was assessed with a cast breast phantom by using a varying number of fiducial crosses (Fig 1). Ten crosses were drawn on this cast and labeled alphabetically, with eight around the perimeter of the breast and two within; all acted as potential fiducial markers. Ten laser surface scans were obtained of the cast. All 10 fiducial crosses were located by using the tracked 3D pointer 10 times and were recorded with the research software as registration points. Each set of registration points could be registered to each laser surface, thus giving 100 pairs of potentially coregisterable data.
View larger version (101K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 1: Laser image of the breast phantom shows alphabetically labeled fiducial mark-ers a–j. On this image, all fiducial points have been coregistered with one another. Black dots represent the laser position of the center of the crosses. White dots were recorded by the US system and were subsequently coregistered to the laser image.
|
|
For each pair of data, every possible combination of groups of three fiducial crosses was considered. For each group, coregistration was performed by using these three points only. The registration accuracy was then assessed by comparing the predicted location of the remaining fiducial crosses that were not used in the registration process with their actual location. For example, if fiducial crosses a–c only were used for coregistration, the accuracy of coregistration was assessed by comparing the predicted locations of points d–j with their actual positions on the laser image (Fig 1). This entire process was repeated by using four to nine points for coregistration, each time only using the remaining points to assess the coregistration process by two investigators (G.M.T., C.E.C.). This exercise produced a series of error measurements for using a variable number of crosses for the coregistration process.
Clinical Study
Data acquisition.—Formal ethical approval for the study was granted by the local research ethics committee. All women recruited to the study gave written consent for participation. From January to April 2004, all patients who received a diagnosis of breast carcinoma with either palpable or impalpable lesions suitable for breast-conserving surgery were invited to participate in the study. Patients whose lesions could not be defined by using US were excluded from the study. Twenty-five patients with a mean age of 57 years (range, 39–73 years) were recruited for the study. Thirteen women had palpable lesions, and 12 had impalpable lesions.
Five days prior to surgery, during the preadmission clinic attendance, the US and laser data were obtained. The patients lay supine with their ipsilateral arm abducted, thus approximating their position at the time of surgery. The patients were asked to remain as still as possible throughout the duration of data acquisition. The tumor was identified with conventional two-dimensional US by either of two breast radiologists (R.S. and P.B., with 7 and 16 years of breast US experience, respectively). After the approximate position of the tumor was identified, six red crosses were marked on the overlying skin. These crosses acted as fiducials for the coregistration process described before.
By using the tracked 3D pointer, the positions of the crosses were recorded to the 800-MHz personal computer equipped with the research software and into the 3D US coordinate system. The 3D US data set was acquired from a simple linear sweep of the transducer over the area of interest. Large amounts of acoustic gel were used in an attempt to reduce the amount of breast tissue compression and contour distortion caused by the transducer. The data set was recorded to the personal computer for later analysis. Residual acoustic gel was wiped from the breast surface. The tripod-mounted laser scanner, positioned approximately 1.5 m away from the patient, was used to acquire the surface contour data of the affected breast to include the fiducial crosses, which were visible on the digital photograph that was simultaneously acquired by one investigator (C.E.C.). With this step, the data acquisition, with an average performance time of 10 minutes (range, 5–12 minutes), was completed.
Data analysis, coregistration, and 3D image production.—Data analysis, coregistration, and the production of a 3D image of the tumor within the breast were performed by one radiologist (C.J.C.C., with 6 years of conventional US and 2 years of 3D US experience). The laser surface contour data with the accompanying photograph was imported into the personal computer equipped with the research software. The recorded position of the fiducial crosses was aligned with the position of the crosses seen on the laser-produced photograph, thereby coregistering the 3D US data set with the laser surface contour data (Fig 2). The success of coregistration was assessed and assigned a grade according to how well the skin surface represented by the US data registered with the surface of the laser data. An objective measurement was made between the two representative skin surfaces from an image of the coregistered data by using the research software. The classification was as follows: grade 1, skin on the US image was aligned with the laser image surface; grade 2, skin at US was within 1 cm from the laser image surface; and grade 3, skin at US was more than 1 cm away from the laser image surface.
View larger version (127K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2a: (a) Photograph of breast in 52-year-old patient 7 acquired simultaneously with the laser image shows that the coregistration process requires the recorded registration points (alphabetically annotated points a–f) to be aligned against the appropriate red fiducial crosses. (b) Laser image representing the surface contour of the breast in same patient shows good registration with close approximation of the coregistered laser-recorded position of the fiducials (red dots) with the research software–recorded position of the fiducials (white dots), corresponding to alphabetically annotated points in a.
|
|
View larger version (94K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 2b: (a) Photograph of breast in 52-year-old patient 7 acquired simultaneously with the laser image shows that the coregistration process requires the recorded registration points (alphabetically annotated points a–f) to be aligned against the appropriate red fiducial crosses. (b) Laser image representing the surface contour of the breast in same patient shows good registration with close approximation of the coregistered laser-recorded position of the fiducials (red dots) with the research software–recorded position of the fiducials (white dots), corresponding to alphabetically annotated points in a.
|
|
The 3D US data set was reviewed at the personal computer. The tumor was manually segmented from the data set, and a series of outlines representing the tumor was produced (Fig 3). These outlines were around the hypoechoic component of the tumor. An estimate of the deep surface of the tumor was made in the event of any posterior acoustic hyporeflectivity. A surface was interpolated from these outlines, thus producing a surface-rendered image of the tumor (15) registered with the laser image. The 3D image produced was transferred to a tablet personal computer (Compaq tc1000; Hewlett-Packard, Woking, England). The time for completion of data analysis, coregistration, and 3D image production was an average of 15 minutes (range, 12–17 minutes). The tablet personal computer was taken to the operating room on the morning of the patient's surgery. A software program was written for the tablet personal computer that enabled the surgeon to dynamically rotate the image in three dimensions (Fig 4). The transparency of the skin could be increased or decreased, thereby varying the definition of the tumor below the surface of the skin (Fig 4). A spotlight tool could be applied that highlights the area of skin closest to the tumor (Fig 4).
View larger version (15K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 3: A, US data set acquired with research software is represented by a series of white soccer-post shapes. The position of the fiducials, recorded as registration points a–f, is shown in relation to the data set, with the first registration point a lying approximately over the scanned area. B, Representative sagittal oblique US image in 52-year-old patient 7 shows tumor (2.1 x 1.2 cm), which has been manually segmented (outlined in red). C, US data set shows series of segmented outlines in red in relation to the registration points.
|
|
View larger version (83K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 4: A–C, Images in 52-year-old patient 7 show tumor in relation to the breast surface viewed dynamically from any angle on a personal computer in the operating room. D, Tumor acts as a light source akin to a radioactive source as used in ROLL and indicates the skin surface closest to the tumor. The red crosses are the remains of the fiducial marks used in the coregistration process.
|
|
Assessment of the 3D image as a localization technique.—In all patients, treatment and surgery were performed according to standard practice. As part of an existing study, the impalpable lesions were localized with a hook wire, with ROLL, or with both. Prior to making the incision, the surgeon (A.D.P.) planned the surgery by marking on the skin a solid circle over the estimated position of the tumor, with the anticipated incision site and excision margin marked as a black line and a circle, respectively; this plan was referred to as plan 1. The surgeon then viewed the 3D image at the tablet personal computer. The image could be orientated to equate to the position of the patient lying on the operating table. After viewing the 3D image, the surgeon planned the surgery again and repeated the marks on the skin in red ink; this plan was referred to as plan 2.
A digital photograph was taken of the breast to record the position of the actual surgical incision (plan 1 in black marks) and the position of the theoretic revised plan for surgery (plan 2 in red marks) that was based on the 3D image. The distance between the center of the solid black and the solid red circles was documented. The surgery was subsequently performed according to plan 1. The results were interpreted according to the relationship of actual surgical plan 1 to the theoretic plan 2 that was based on the 3D image, and patients were classified into three groups. Group A included patients in whom the position of plan 2 was the same as plan 1. In group A, it was assumed that the technique had been equivalent to conventional methods of localization. Group B included patients in whom the position of plan 2 was up to or equal to 1 cm of plan 1. Group C included patients in whom the position of plan 2 differed from plan 1 by more than 1 cm.
Surgical and histopathologic outcome.—In all patients, the histopathologic results from the excised specimen were obtained. It is current local policy for reexcision to be performed, if clinically appropriate, in all patients in whom the radial excision margin is less than 5 mm. In this study, the localization technique relies on US-detectable tumor. Any ductal carcinoma in situ component may or may not be seen at US, and therefore the localization technique was assessed on the basis of its capability to localize invasive tumor and not ductal carcinoma in situ. For any patients with widespread ductal carcinoma in situ who required further surgery, the localization technique was assessed on the basis of the margin around the invasive tumor.
Breast volume assessment.—Breast volume was anticipated to influence the outcome of the localization technique. Hence, breast volume was assessed from the preoperative mammograms (16) by a radiologist (R.S., with 7 years of breast radiologic experience) who was blinded to the results of the localization technique. The volume was calculated from the formula 
dmamfh (16), where h is the distance from the nipple to the chest wall just below the pectoral fold and dmamf is the distance between the supramammary (axillary) and inframammary folds measured on the mediolateral oblique mammogram.
Statistical Analysis
Validation studies.—The effect of using different colored fiducials on the reconstructed surface was compared with that of using no marks on a surface by using the F test (http://davidmlane.com/hyperstat/F_table.html). To assess the accuracy of the coregistration process by using different numbers of fiducial points, the worst-case error was recorded and the accuracy with a 95% confidence level was calculated (Matlab 6.5, June 2002; MathWorks, Natick, Mass).
Clinical studies.—The effect of the 3D localization technique compared with existing localization methods was assessed in women with breast volume of 1000 mL or greater and in women with breast volume less than 1000 mL by using the Fisher exact test. A difference with a P value of less than .05 was considered statistically significant.
|
RESULTS
|
|---|
Validation Results
The reconstructed skin surface was affected by the color of the pen used to place a cross (fiducial). Neither the pink crosses (P > .99) nor the red crosses (P = .41) had a significant effect on the surface, as shown with the F test (Table 1). Because red is much more visible to the naked eye, all subsequent clinical studies were performed with red crosses.
View this table:
[in this window]
[in a new window]
|
Table 1. Effect of Placement of Crosses with Varying Pen Colors Compared with Placement of No Marks on Reconstructed Skin Surface
|
|
The accuracy of coregistration varied according to the number of fiducial points (Table 2). In practice, these results are highly pessimistic, since the points used to assess registration errors often lay well outside of the points used for registration itself (eg, if registration points a–c were used for coregistration, the registration error assessed in point f some distance away would be greater than the registration error in point i (Fig 1). In subsequent clinical studies, fiducial marks were placed surrounding the area of interest. It was found that coregistration accuracy could be kept to within 3 mm by using at least five fiducial points (with a 95% confidence level) (Table 2).
3D Image Production
All lesions were definable by using US but with varying degrees of clarity. In the group of patients with palpable lesions (n = 13), all patients had clearly defined tumors, but in the group of patients with impalpable lesions (n = 12), one patient had a poorly defined tumor and one patient had a tumor for which adequate resolution was impossible to achieve without applying a considerable amount of compression with the transducer (Fig 5). By using our coregistration classification system, 11 registrations were classified as grade 1 (Fig 6), five were classified as grade 2, and nine were classified as grade 3. Grade 3 registrations were due to compression (n = 8) and patient movement during data acquisition (n = 1).
View larger version (104K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5a: Transverse 3D US images in 56-year-old patient 21 illustrate the typical change in image definition at the depth when compression induced by the transducer is (a) increased or (b) decreased. Arrow points to hypoechoic tumor.
|
|
View larger version (102K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 5b: Transverse 3D US images in 56-year-old patient 21 illustrate the typical change in image definition at the depth when compression induced by the transducer is (a) increased or (b) decreased. Arrow points to hypoechoic tumor.
|
|
View larger version (103K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6a: (a) Full data set coregistered with laser image in 52-year-old patient 7. Grade 1 registration is demonstrated; the top of the images within the data set (ie, the skin surface) closely align with the skin surface of the laser image. (b) Position of segmented tumor alone (red outlines) is shown in same patient. The original US data set is represented by the white soccer-post shapes. The fiducial markers at points a –f used for coregistering the US data (white points) with the laser image (red points) are shown.
|
|
View larger version (105K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 6b: (a) Full data set coregistered with laser image in 52-year-old patient 7. Grade 1 registration is demonstrated; the top of the images within the data set (ie, the skin surface) closely align with the skin surface of the laser image. (b) Position of segmented tumor alone (red outlines) is shown in same patient. The original US data set is represented by the white soccer-post shapes. The fiducial markers at points a –f used for coregistering the US data (white points) with the laser image (red points) are shown.
|
|
Three-dimensional Localization Technique and Surgical and Histopathologic Outcomes
Details of localization, surgery, and histopathologic outcome in 25 patients are presented in Table 3. Three patients required further surgery for multifocal disease (n = 1) and widespread ductal carcinoma in situ (n = 2).
View this table:
[in this window]
[in a new window]
|
Table 3. Details of Localization, Surgery, and Histopathologic Outcome in All 25 Patients: Potential Effect of 3D Localization Technique
|
|
In 12 of 25 patients (group A), the position of plan 1 was identical to the position of plan 2 (Fig 7). In six of 25 patients (group B), the position of plan 2 differed from that of plan 1 by up to 1 cm (Fig 8). However, in five of six patients from group B, this difference in position was noted to be in the same radial direction as the position of cavity shavings obtained at the time of surgery or in the same direction as the minimal radial excision margin (Table 3). For example, in patient 16 (Fig 8), the position of plan 2 lies just lateral to that of plan 1. Histopathologic correlation shows that the smallest clear excision margin was on the lateral side (4 mm), with a generous medial margin of 20 mm.
View larger version (153K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 7: Marked breast in 62-year-old patient 6 shows position of plan 1 (conventional technique) marked in black ink and position of plan 2 (3D US localization method) marked in red ink. It was assumed that if the position of plan 2 was the same as plan 1, as in this patient, the 3D image localization technique was equivalent to the conventional localization method.
|
|
View larger version (142K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 8: Marked breast in 65-year-old patient 16 shows that although position of plan 2 (red ink) differed from position of plan 1 (black ink), the corresponding histologic findings suggested that, had the surgery been performed according to plan 2, a satisfactory outcome would have been achieved.
|
|
In patient 13 from group B, the original diagnostic US and core biopsy findings had revealed just one tumor. At the time of the 3D US data acquisition, two tumors were identified. The histopathologic outcome revealed three tumors within the excised specimen. If multifocality had been known at diagnosis, this patient would have received a recommendation for mastectomy in the first instance. In this patient, the 3D localization method was, therefore, considered satisfactory even though the final outcome was a mastectomy. It was assumed therefore that the technique had been equivalent to conventional methods in patients from group B.
In seven of 25 patients (group C), the position of plan 2 differed from plan 1 by more than 1 cm (Fig 9). One of these patients (patient 24) is known to have moved during data acquisition. This patient was elderly, had lumbar vertebral osteoarthritis, and had difficulty lying flat. In view of her continuing discomfort, the registration process unfortunately could not be repeated. In four patients, the breast volumes were very large or above average (Fig 9). In the two remaining patients in whom the position of plan 2 differed from plan 1 by more than 1 cm, neither the breast volume nor the mentioned patient movement accounted for the result. It was noted, however, that in one patient there was extreme laxity to the breast tissue. In the other patient, with an upper outer quadrant lesion, it was clear from the photograph taken of the surgical plans in the operating room, that the plans had been determined after axillary clearance had been performed.
View larger version (153K):
[in this window]
[in a new window]
[Download PPT slide]
|
Figure 9: Marked breast in 56-year-old patient 21 shows that the perceived position of the tumor according to the 3D image (red ink) differed from the perceived position according to the conventional localization method (black ink). When the tumor lies deep within a large breast, it is difficult to assess whether the 3D localization method would have been successful.
|
|
To summarize the results (Table 3), the technique appears to have been successful (defined as either an improvement in localization compared with existing localization methods or localization equivalent to existing methods) in 72% (18 of 25) of patients (groups A and B) and unsuccessful (defined as less accurate than existing localization methods) in 28% (seven of 25) (group C).
Breast Volumes
The results of the breast volume assessment were available for 23 of the patients (Table 3). Volume calculation was not possible for one patient with breast implants and in another patient in whom the mammograms were not available. The success rate for the 3D method is slightly higher for women with a smaller breast volume than for women with a larger breast volume, but the difference was not significant (P = .64, Fisher exact test).
|
DISCUSSION
|
|---|
The method we describe is noninvasive, is safe, and does not involve ionizing radiation. Although the total time for production of the 3D image of the tumor within the breast may be longer than the time taken for conventional techniques of localization, the data can be acquired a few days prior to surgery. To our knowledge, breast tumor localization with 3D US and 3D surface contour mapping is the only localization method in which the surgeon can view not only the location of the tumor within the breast but also the 3D tumor shape. Spatial orientation of the tumor with respect to the nipple may illustrate ductal extension of tumor, and this factor may have an influence on the shape and volume of the excised specimen.
The validation work results suggested that the coregistration of a US data set to a laser surface was accurate to within 3 mm, provided at least five fiducial points were used for coregistration. In each of the clinical studies, six fiducial points were used. With the phantom work, we could not take into account errors introduced by patient respiration and other movement, nor errors from breast compression by the transducer.
In this clinical feasibility study, we compared a noninvasive method of tumor localization with current methods of localization, which were clinical palpation in the patients with palpable lesions and a combination of ROLL, hook-wire, and skin-marking in the patients with impalpable lesions. We assumed that if the position of plan 2 (with use of the 3D image localization method) was the same as plan 1 (with use of conventional methods of localization) or the position was within 1 cm of plan 1, taking into account cavity shavings and the histopathologic results, the 3D image localization method was as good as localization with conventional techniques (72%, 18 of 25 patients). It is possible that, in five of the six patients from group B, if surgery had been performed according to the 3D image localization method, a more acceptable histopathologic outcome may have been achieved with more evenly distributed clear margins.
In seven patients (group C), it was difficult to prove or disprove that the technique would have resulted in a satisfactory outcome although the technique was described as less successful than conventional techniques. Patient movement, breast volume, and breast laxity are confounding factors in this group. It is difficult to evaluate the potential surgical outcome by using this localization technique in the larger-breasted patients. There must be several potentially successful surgical approaches for a deeply placed tumor within a large-volume breast. Therefore, although plan 2 was discrepant from plan 1 in this group, it is possible that a successful surgical outcome would have been achieved had the surgeon performed surgery according to plan 2.
Breast volume is likely to influence the outcome of this localization method for two reasons. First, image definition diminishes with depth, thus resulting in the need for compression. Compression will distort the breast contour and result in a poor (grade 3) coregistration of US data with laser data. Although breast compression is likely to cause movement of the tumor into the same plane as the longitudinal axis of the excision cylinder, this movement into the same plane cannot always be assumed and in some cases the tumor may be shifted radially (ie, superiorly, inferiorly, medially or laterally). This factor may explain why the technique appeared to work well in some of the patients with larger breasts and not in others. A high-frequency transducer was chosen in this study to provide morphologic detail as well as spatial resolution. In retrospect, a lower-frequency transducer that provides greater depth penetration may have been more suitable in the larger-breasted patients.
Second, a small object deep within a larger object may present perceptual difficulties for the surgeon. The ability of the surgeon to translate the 3D position of an object from an image to its actual position in vivo has not been assessed but is undoubtedly fundamental to this new technique. With the method, one assumes that the contour of the breast during data acquisition will be similar to the contour of the breast on the operating table. Any changes in contour due to breast laxity or axillary clearance performed prior to surgical planning may affect the success of the technique. Semipermanent skin marks visible on the 3D image and on the breast at the time of surgery may improve spatial orientation.
There are a number of limitations to this study in a small number of patients. The accuracy of the coregistration stage of the clinical study was assessed by correlating the position of the skin from the US image with the position of the surface contour from the laser image. With this part of the method, only one plane of potential misregistration is assessed. Although the commonest causes of misregistration affect this plane (ie, compression and respiration), misregistration caused by patient movement for imaging in the mediolateral or the superoinferior plane has not been assessed. The recording of a seventh visible landmark (eg, the nipple) by the 3D pointer at the end of the US data acquisition could be introduced to assess this potential error. The accuracy and repeatability of the process of manual segmentation was not formally assessed in this study, and we are currently comparing the interobserver and intraobserver variability. Outlining around the hypoechoic component of the tumor is likely to cause underestimation of the tumor size. Because many breast tumors have poorly defined margins on US images, defining the hypoechoic tumor margin was a consistent approach. Because the primary objective of the study was tumor location, rather than tumor size or outline, this potential variation was not assessed. In the assessment of the localization technique, it is possible that some bias may have been introduced; the surgeon had already planned the actual surgery prior to viewing the 3D image and this factor may have influenced the interpretation of the 3D image. Ideally, half of the patients could have been planned in reverse order (ie, allowing the surgeon to view the 3D image prior to planning the actual surgery). This would, however, raise the ethical issue of whether the surgeon could have been misled by a localization technique that had not undergone any form of clinical evaluation.
This technique is most likely to be suitable in localization of impalpable tumors within small to average breasts. It is imperative that the patient is able to lie supine and immobile. Conventional techniques are likely to be superior in the larger-breasted patient, thus providing the necessary physical guidance rather than virtual localization.
The results of this feasibility study are promising. To be implemented, the method would need to be formally examined in the setting of a randomized trial in which this experimental method of localization would be compared with existing localization techniques. It is anticipated that, with patient selection, rigorous patient alignment during data acquisition, skin marking, and training of surgeons in the use of the 3D images, this method for localization of impalpable lesions would be accurate and acceptable to the patient and surgeon alike.
|
ADVANCE IN KNOWLEDGE
|
|---|
- It is possible to produce a three-dimensional reconstruction of the preoperative breast that shows the position and shape of a tumor in relation to the skin surface by using three-dimensional US and surface contour mapping.
|
IMPLICATION FOR PATIENT CARE
|
|---|
- The coregistration procedure we describe is a noninvasive method of breast tumor localization that can be performed prior to breast-conserving surgery.
|
FOOTNOTES
|
|---|
Abbreviations: ROLL = radioactive occult lesion localization • 3D = three-dimensional
Guarantor of integrity of entire study, C.J.C.C.; study concepts/study design or data acquisition or data analysis/interpretation, all authors; manuscript drafting or manuscript revision for important intellectual content, all authors; manuscript final version approval, all authors; literature research, C.J.C.C.; clinical studies, C.J.C.C., C.E.C., G.M.T., A.D.P., P.B., R.S.; experimental studies, C.J.C.C., C.E.C., G.M.T., A.H.G., R.W.P.; statistical analysis, C.J.C.C., C.E.C., G.M.T.; and manuscript editing, C.J.C.C., A.D.P., P.B., R.S.
Authors stated no financial relationship to disclose.
|
References
|
|---|
- Kopans DB, Swann CA. Preoperative imaging-guided needle placement and localization of clinically occult breast lesions. AJR Am J Roentgenol 1989;152:1–9. [Free Full Text]
- Homer MJ. Localization of nonpalpable breast lesions: technical aspects and analysis of 80 cases. AJR Am J Roentgenol 1983;140:807–811. [Free Full Text]
- Homer MJ. Percutaneous localization of breast lesions: experience with the Frank Breast Biopsy Guide. J Can Assoc Radiol 1979;30:238–241. [Medline]
- Gisvold JJ, Martin JK Jr. Prebiopsy localization of nonpalpable breast lesions. AJR Am J Roentgenol 1984;143:477–481. [Abstract/Free Full Text]
- Bronstein AD, Kilcoyne RF, Moe RE. Complications of needle localization of foreign bodies and nonpalpable breast lesions. Arch Surg 1988;123:775–779. [Abstract/Free Full Text]
- Bristol JB, Jones PA. Transgression of localizing wire into the pleural cavity prior to mammography. Br J Radiol 1981;54:139–140. [Abstract/Free Full Text]
- Tykka H, Castren-Persons M, Sjoblom SM, Roiha M. Pneumothorax caused by hooked wire localisation of an impalpable breast lesion detected by mammography. Breast 1993;2:52–53. [CrossRef]
- Homer MJ. Transection of the localization hooked wire during breast biopsy. AJR Am J Roentgenol 1983;141:929–930. [Free Full Text]
- della Rovere GQ, Benson JR, Morgan M, Warren R, Patel A. Localization of impalpable breast lesions: a surgical approach. Eur J Surg Oncol 1996;22:478–482.
- De Cicco C, Pizzamiglio M, Trifiro G, et al. Radioguided occult lesion localisation (ROLL) and surgical biopsy in breast cancer: technical aspects. Q J Nucl Med 2002;46:145–151.
- Gray RJ, Salud C, Nguyen K, et al. Randomized prospective evaluation of a novel technique for biopsy or lumpectomy of nonpalpable breast lesions: radioactive seed versus wire localization. Ann Surg Oncol 2001;8:711–715. [Abstract/Free Full Text]
- Snider HC Jr, Morrison DG. Intraoperative ultrasound localization of nonpalpable breast lesions. Ann Surg Oncol 1999;6:308–314. [Abstract]
- Rahusen FD, Bremers AJ, Fabry HF, van Amerongen AH, Boom RP, Meijer S. Ultrasound-guided lumpectomy of nonpalpable breast cancer versus wire-guided resection: a randomized clinical trial. Ann Surg Oncol 2002;9:994–998. [Abstract/Free Full Text]
- Treece GM, Gee AH, Prager RW, Cash CJ, Berman LH. High-definition freehand 3-D ultrasound. Ultrasound Med Biol 2003;29:529–546. [CrossRef][Medline]
- Treece GM, Prager RW, Gee AH, Berman L. Surface interpolation from sparse cross sections using region correspondence. IEEE Trans Med Imaging 2000;19:1106–1114. [CrossRef][Medline]
- Katariya RN, Forrest AP, Gravelle IH. Breast volumes in cancer of the breast. Br J Cancer 1974;29:270–273. [Medline]
TOP
ABSTRACT
INTRODUCTION
