Breast cancer is the most common form of cancer among women in Japan. The bone is the first metastasis site in 26–50% of breast cancer patients . Autopsies of breast cancer patients have found bone metastases in about 65% of cases . Therefore, it is essential to be able to diagnose bone metastasis accurately.
Tools such as computed tomography (CT), MRI, scintigraphy, and PET-CT are used to diagnose bone metastasis. Bone scintigraphy in particular is often used in routine clinical practice because it can evaluate all the bones of the body at once. Single-photon emission computed tomography (SPECT) has been used to improve spatial resolution.
For bone metastasis of breast cancer, SPECT is reported to have a sensitivity of 87–92% and a specificity of 91–93% . However, bone scintigraphy reportedly has a sensitivity of 62–100% and a specificity of 78–100% when diagnosing bone metastasis of breast cancer . Other studies have reported a sensitivity of 33–100% (median 98%) and specificity of 82–100% (median 96.5%) , as well as false positive rates of 10–22% and a false negative rate of 10% . This is thought to be due to the low spatial resolution of bone scintigraphy and the tendency for the interpreter’s subjectivity to affect visual evaluations.
To address this, computer-aided diagnostic (CAD) systems have been developed for use with bone scintigraphy. In Japan, Fujifilm RI Pharma and EXINI Diagnostics jointly developed the BONENAVI system, a CAD system that can be applied to bone scintigraphy using 99mTc-MDP to obtain greater objectivity. It was approved as BONENAVI BSI on 27 May 2015. There are two versions of the BONENAVI system. BONENAVI version 1 is loaded with a database of Japanese patients from a single institution and is reported to have a sensitivity of 90%, specificity of 81%, and a diagnostic accuracy of 82% . BONENAVI version 2 is loaded with a database of Japanese patients from multiple institutions and can perform separate analyses by type of cancer, such as for prostate or breast cancer . The area under the curve (AUC) for breast cancer is 0.910 for BONENAVI version 1 and 0.924 for BONENAVI version 2, which is not a statistically significant difference .
However, there have been no reports on the sensitivity of BONENAVI version 1 and BONENAVI version 2 for bone metastasis of primary breast cancer, or any direct comparisons of these versions. The objective of this study was to compare the sensitivity of BONENAVI version 1 and BONENAVI version 2 for bone metastasis of primary breast cancer.
The subjects were selected from patients who underwent bone scintigraphy using 99mTc-MDP at St. Marianna Medical University Hospital from January 2006 to November 2015. A period during this time frame in which raw data from BONENAVI version 1 (switched to BONENAVI version 2 in 2015) were available and saved on interpretation terminals and that included sufficient cases (at least 50) for statistical analysis was examined retrospectively. Ultimately, we examined patients diagnosed with bone metastasis of primary breast cancer based on CT, MRI, clinical examination, and bone scintigraphy findings, who underwent bone scintigraphy using 99mTc-MDP at St. Marianna Medical University Hospital from January 2012 to November 2014.
Bone scintigraphy analysis
Bone scintigraphy was performed by intravenously administering 99mTc-MDP (Fujifilm RI Pharma Co., Tokyo, Japan) 370–925 MBq; 3 hours later, horizontal images (anterior and posterior) were taken using a scintillation camera (ECAM and GX7200, Toshiba Corp., Tokyo, Japan). The ECAM imaging rate was 18 cm/min and the image matrix size was 256 × 1024. The GX7200 imaging rate was 17.5 cm/min and image matrix was 256 × 1024. Both devices used a parallel multihole collimator. The data acquisition window was 10% of both ends of the 140 KeV gamma ray emitted by 99mTc-MDP.
Both BONENAVI version 1 (Fujifilm RI Pharma) and BONENAVI version 2 (Fujifilm RI Pharma) were used to calculate artificial neural networks (ANN), bone scan index (BSI), and hot-spot values from the bone scintigraphy data (Fig. 1).
The ANN, BSI, and hot-spot values of BONENAVI version 1 and BONENAVI version 2 were compared using t-tests. ANN ≥ 0.5 was considered positive for bone metastasis . The sensitivity of BONENAVI version 1 and BONENAVI version 2 was calculated and compared using McNemar’s text. EZR statistics software developed by Jichi Medical University’s Saitama Medical Center was used. P < 0.05 was considered to indicate a statistically significant difference.
This study was conducted with the approval of the St. Mariana University School of Medicine ethics committee (approval no. 2977).
56 continuous cases were selected. They were diagnosed with bone metastasis of primary breast cancer based on CT, MRI, and clinical findings, and visual assessments of their bone scintigraphy findings were not considered contradictory with this diagnosis. All patients were women. Their mean age was 59 ± 12.7 years.
Comparison of artificial neural network, bone scan index, and hot-spot values
With BONENAVI version 1, mean ANN was 0.73 ± 0.29, BSI was 1.47 ± 1.85, and the hot-spot value was 12.4 ± 12.5. With BONENAVI version 2, mean ANN was 0.86 ± 0.19, BSI was 1.53 ± 2.09, and hot spot was 12.9 ± 15.6.
The ANN value was significantly higher with BONENAVI version 2 than with BONENAVI version 1 (P < 0.01). BSI and hot-spot values were not significantly different between BONENAVI version 1 and BONENAVI version 2 (BSI P = 0.88 and hot-spot P = 0.86; Table 1).
We show two clinical cases, case 1: both BONENAVI version 1 and BONENAVI version 2 were positive diagnosis (both ANN value was above 0.5) (Fig. 2); and case 2: BONENAVI version 1 were negative diagnosis and BONENAVI VERSION 2 were positive diagnosis (BONENAVI version 1 ANN value was under 0.5 and BONENAVI version 2 ANN value was over 0.5) (Fig. 3).
Comparison of sensitivity
Sensitivity was 76.8% (43/56) with BONENAVI version 1 and 94.6% (53/56) with BONENAVI version 2. 11 of the 13 patients who exhibited ANN < 0.5 with BONENAVI version 1 exhibited ANN > 0.5 with BONENAVI version 2, with the other 2 patients having ANN < 0.5. 1 of the 43 patients that exhibited ANN > 0.5 with BONENAVI version 1 exhibited ANN < 0.5 with BONENAVI version 2. Using McNemar’s test, the sensitivity of BONENAVI version 2 based on ANN was significantly better than that of BONENAVI version 1 (P < 0.01).
Comparison of sensitivity between bone scan index ≤ 1 and bone scan index > 1 groups
23 of the 33 patients who exhibited BSI ≤ 1 with BONENAVI version 1 exhibited ANN > 0.5, yielding a sensitivity of 69.7% (23/33). 9 of these 10 patients, who exhibited ANN < 0.5 with BONENAVI version 1 exhibited ANN > 0.5 with BONENAVI version 2. Using McNemar’s test, a significant difference in sensitivity based on BSI was observed between BONENAVI version 2 and BONENAVI version 1 (P = 0.027).
20 of the 23 patients who exhibited BSI > 1 with BONENAVI version 1 exhibited ANN > 0.5, indicating a sensitivity of 87.0% (20/23). 2 of the remaining 3 patients who exhibited ANN < 0.5 with BONENAVI version 1 exhibited ANN > 0.5 with BONENAVI version 2. Based on McNemar’s test, there was no significant difference in sensitivity between BONENAVI version 2 and BONENAVI version 1 (P = 0.48).
Bone scintigraphy is widely used in routine clinical practice to diagnose bone metastasis. However, diagnosis using bone scintigraphy is based on the locations and patterns of radiopharmaceutical concentrations, which relies on the interpreter’s experience and is thus easily influenced by subjectivity. CAD systems have been developed to improve the objectivity of diagnosis with bone scintigraphy. BONENAVI calculates three parameters: ANN, BSI, and hot-spot values. ANN is an indicator that expresses the probability of an abnormal concentration of radionuclide. The probability of an abnormal concentration in the bone in question is automatically calculated as a continuous value from 0 to 1, with values ≥0.5 indicating bone metastasis. The BSI expresses the proportion of areas that are considered highly likely to have bone metastases within the entire skeleton. The hot-spot value is the number of sites considered highly likely to be bone metastases BONENAVI uses these three parameters as supplemental indicators during diagnosis.
There have been several reports on CAD systems for use with bone scintigraphy. The sensitivity and specificity of CAD systems using horizontal images from bone scintigraphy were previously reported as 90% (19/21) and 89% (34/38), respectively . Another study that examined diagnostic performance with and without a CAD system found no change in specificity, although sensitivity increased from 78 to 88% when a CAD system was used . However, directly applying information from databases based on Europeans and Americans could lead to an increase in false-positives due to differences in their physiques and other factors. Therefore, BONENAVI version 1 was developed using a database of more than 900 Japanese patients. When using the BONENAVI version 1 to examine Asian cases, the sensitivity, specificity, and diagnostic accuracy were 90%, 81%, and 82%, respectively, while when using the European–American database to examine the same cases, these values were only 83%, 57%, and 61%, respectively . Yet, the database for BONENAVI version 1 did not distinguish between men and women or between cancers and was constructed with data from only a single institution. Therefore, BONENAVI version 2 was developed using additional cases from multiple institutions and allowed analyses to be conducted by sex and type of cancer . The areas under the curves for breast cancer were 0.910 and 0.932 with BONENAVI version 1 and BONENAVI version 2, respectively, which were not significantly different . The sensitivity for bone metastasis of breast cancer was previously reported as 82% (37/45) with BONENAVI version 2 . However, in the present study, the sensitivity was 76.8% [43/56] with BONENAVI version 1 and 94.6% [53/56] with BONENAVI version 2, that is, sensitivity was significantly better with the latter version. Furthermore, when BSI ≤ 1, sensitivity was 69.7% (23/33) with BONENAVI version 1, although 9 of these 10 patients who also exhibited ANN < 0.5 exhibited ANN > 0.5 with BONENAVI version 2, which was significantly different. In addition, ANN was significantly different between BONENAVI version 1 and BONENAVI version 2 at 0.73 ± 0.29 and 0.86 ± 0.19, respectively (P < 0.01), although no significant differences were observed for BSI or hot-spot values. These findings indicate that BONENAVI version 2 has better sensitivity than previously reported, and that while BSI probably affected ANN values with BONENAVI version 1, this effect seems to have been addressed in BONENAVI version 2. The reasons for this are unclear, although possible reasons include the timing of bone metastasis and the number of cases in the loaded database.
The databases of BONENAVI version 1 and BONENAVI version 2 were not limited to cases of bone metastasis from primary cancer. However, in addition to chemotherapy and hormone therapy, drugs intended to reduce bone-related adverse effects, such as zoledronic acid and denosumab, are used in cases of bone metastasis. Like 99mTc-MDP, zoledronic acid is also a bisphosphonate; these agents may therefore contend with each other, affecting concentrations and diagnostic precision. In addition, because denosumab is a RANKLE antibody that changes bone metabolism, it may also have an impact on diagnostic precision. When chemotherapy is initiated for bone metastasis, zoledronic acid and denosumab may modify the concentrations observed with bone scintigraphy, which could affect the use of neural networks in diagnosing the presence or absence of bone metastases.
Moreover, for BONENAVI version 2, a database was created using a large number of cases from multiple institutions. This likely increased the spectrum of bone metastases that could be recognized, thereby improving sensitivity. In other words, while BSI and hot-spot values did not differ between BONENAVI version 1 and BONENAVI version 2, the significant improvement in ANN values with BONENAVI version 2 resulted in better sensitivity in the diagnosis of bone metastases, because in addition to the amount of bone metastases (i.e. BSI), the neural network was able to recognize factors such as distribution.
This study had several limitations. In addition to including a limited number of cases, the investigation was conducted at a single institution. Going forward, further investigations should include more cases and at multiple institutions.
BONENAVI version 2 had better sensitivity for detecting bone metastasis of primary breast cancer than BONENAVI version 1.
Yukinori Okada and Shoichiro Matsushita had the original idea and collected and analysed the data. Yasuyuki Kojima checked and suggested the clinical diagnosis. Yasuo Nakajima, Keiichiro Yamaguchi, Itsuko Okuda, and Koichiro Tsugawa suggested the data analysis methods and checked the data analysis results.
Fujifilm RI Pharma provided JPY 500,000 in research funding for other research but, in this research, we had no conflict of interest.
Conflicts of interest
There are no conflicts of interest.
1. Makita M, Iwase T, Tada T, Takahashi K, Tada K, Nishimura S, et al. The site and timing of the first recurrence of breast cancer. Jpn J Breast Cancer. 2004; 19:343–351
2. Hamaoka T, Madewell JE, Podoloff DA, Hortobagyi GN, Ueno NT. Bone
imaging in metastatic breast cancer. J Clin Oncol. 2004; 22:2942–2953
3. Brennan ME, Houssami N. Evaluation of the evidence on staging imaging for detection of asymptomatic distant metastases in newly diagnosed breast cancer. Breast. 2012; 21:112–123
4. Myers RE, Johnston M, Pritchard K, Levine M, Oliver T; Breast Cancer Disease Site Group of the Cancer Care Ontario Practice Guidelines Initiative. Baseline staging tests in primary breast cancer: a practice guideline. CMAJ. 2001; 164:1439–1444
5. Horikoshi H, Kikuchi A, Onoguchi M, Sjöstrand K, Edenbrandt L. Computer-aided diagnosis system for bone
scintigrams from Japanese patients: importance of training database. Ann Nucl Med. 2012; 26:622–626
6. Nakajima K, Nakajima Y, Horikoshi H, Ueno M, Wakabayashi H, Shiga T, et al. Enhanced diagnostic accuracy for quantitative bone
scan using an artificial neural network system: a Japanese multi-center database project. EJNMMI Res. 2013; 3:83
7. Koizumi M, Miyaji N, Murata T, Motegi K, Miwa K, Koyama M, et al. Evaluation of a revised version of computer-assisted diagnosis system, BONENAVI version 2.1.7, for bone
scintigraphy in cancer patients. Ann Nucl Med. 2015; 29:659–665
8. Sadik M, Hamadeh I, Nordblom P, Suurkula M, Höglund P, Ohlsson M, Edenbrandt L. Computer-assisted interpretation of planar whole-body bone
scans. J Nucl Med. 2008; 49:1958–1965
9. Sadik M, Suurkula M, Höglund P, Järund A, Edenbrandt L. Improved classifications of planar whole-body bone
scans using a computer-assisted diagnosis system: a multicenter, multiple-reader, multiple-case study. J Nucl Med. 2009; 50:368–375