White Paper S-Detect Breast 2 RS80A HS70A 2015

Article # WP WP201509-S-Detect / Issue Date 9 Sept, 2015
Evaluation of S-Detect as a screening

tool for Breast Tumors
Phan Thanh Hai Phuong,MD, Jasmine Thanh Xuan, MD, Nguyen Nghiep Van, MD,
Le Dinh Vinh Phuc, MD, Phan Thanh Hai, MD
Medic HCM, Vietnam
“ With its provided accuracy, S-Detect can certainly become a valuable tool in breast
cancer screening, and even also assist beginners and non-experienced radiologists. ”
Introduction
Breast cancer is the most common cancer with high mortality rate in women internationally, including
Vietnamese women. Therefore, an early detection of breast cancer is very crucial for better prognosis.
Among diagnostic modalities, breast ultrasound is a useful screening tool in differentiating benign
and malignant tumors. However, its limitation still depends largely on radiologist’s experience.
Recently, for this reason, computer-aided diagnosis (CAD) systems have been rapidly entering the
radiology mainstream, providing second opinion in assisting image interpretation. Based on the
definitions in mass category of Breast Imaging Reporting and Data System developed by American
College of Radiology 2013 ultrasound lexicon (BI-RADS)1, Samsung Medison developed a CAD
software called S-Detect to help physicians making an accurate decision. It is clearly finding its way
to the clinical application.
In this study, 111 breast lesions were collected and they were reconfirmed of their pathology by
biopsy at Medic Center, Ho Chi Minh City. With these results, S-Detect’s efficacy was evaluated and
the agreement between S-Detect and Conventional breast ultrasound examination was measured.
Method
First, a cross-sectional study was designed and all of the data were collected at ultrasound department
of Medic Center, Ho Chi Minh City. Using sample size determination formula, with an expected
margin of error at w = 0.05 and Pdis = 20/100,000 (prevalence of breast cancer in Vietnamese women
population, according to Nguyen Chan Hung et al2), the minimum size of 73 cases was calculated.
In addition, since observation occurs in different population, the number of cases must be 1.5 to 2
fold the minimum size needed. Therefore, a total number of cases needed were calculated 1.5 times
the minimum (n = 73x1.5 = 110 cases).
Second, from April 2015 to June 2015, data of 111 breast lesions of 111 females patients whose
BI-RADS scores were from 3 to 5 were acquired. Patients who disagreed to be a part of the research
and/or patients with surgical and chemotherapy history were excluded from the study.
Lastly, the data were statistically analyzed using software SPSS 16.0 and p value<0.05 was considered
as significant.
Two radiologists whose experiences are 1 and 15 years in breast US investigated dedicated lesions using
linear probe L3-12A in RS80A with Prestige (Samsung Medison Co., Ltd). Lesions were reported based on
BI-RADS final assessment categories. S-Detect was applied to re-estimate the lesions. The specialized
computer algorithm analyzed based on 6 features – shape, orientation, margin, lesion boundary, echo
pattern, posterior feature. Finally, the system suggested a possibility of being benign or malignant. All
lesions’ final diagnosis were confirmed by biopsy results (FNAC, core biopsy, excisional biopsy).
Results & Discussion

1) Evaluation of S-Detect
Malignant Benign Sum

Possibly malignant 57 19 76
Possibly benign 4 31 35
Table 3.1. S-Detect results compared with biopsy results (n = 111)
A total of 111 malignant lesions were included in series, and 61 were malignant (60 carcinoma, 1
sarcoma). Of these 61 cases, 57 were reported malignant by S-Detect (Figure 1 and 2). The sensitivity
based on these results was 93.4%. The remaining 4 out of 61 malignant cases (6.65%) were
cancerous tumors which software conveyed as benign. They were all subclinical lesions (< 1cm) with
circumscribed border, and no signs of surrounding tissue infiltration.
Figure 1. A right breast mass classified as BI-RADS 5. S-Detect also proposed as possibly malignant. Biopsy revealed sarcoma.
Figure 2. A left breast mass in 48 years old patient, <1cm in diameter, speculated margin, vertical orientation, Halo
sign(+). S-Detect and Conventional US both suggest malignant. Biopsy revealed Ductal Carcinoma in Situ (DCIS).
In addition, 3 of them were satellite lesions nearby large masses, and 1 was an early detected
carcinoma. (Figure 3)
Among 50 benign masses, S-Detect reported correctly 31/50 making its specificity at 62% and
accuracy at 79.3%. The misinterpreted were the cases of fat necrosis, breast abscess, infected simple
cyst, tuberculous mastitis, fibroadenoma, and etc., which show varying hypoechoic to heterogeneous
echo patterns, ill-defined border, vertical oriented and sometimes halo sign (+). Especially for tumors
with ill-defined border or vertical orientated alone, S-Detect usually suggested malignancy. However,
in these cases, the over-diagnosis was necessary to avoid false-negative. However, this software also
has limitation. The existence of microcalcifications and neovascular pattern were not analyzed. Since
they are valuable indicators for malignancy, S-Detect misses an interpreting information.
Figure 3. Satellite lesion with <1cm diameter, well-defined margin, horizontal orientation, halo sign(-). S-Detect
suggested benign. Biopsy revealed Invasive Ductal Carcinoma (IDC)
2) Correlation between S-Detect and Conventional breast ultrasound
Conventional US
Possibly benign Possibly malignant Sum
Possibly malignant 31 4 35
S-Detect Possibly benign 8 68 76
Sum 39 72 111
kappa coefficient 0.76
Table 3.2. Correlation between S-Detect and conventioanl US in diagnosis of breast cancer (n = 111)
To measure the agreement of these two methods, kappa coefficient was calculated at 0.76.
Using cited scale in Table 3.3, the kappa coefficient of 0.76 is in “Substantial agreement” range. As
a result, it’s implied that S-Detect has a high correlation with the alternative method. This supports
S-Detect’s application in daily clinical breast practice in assisting investigation of breast tumors.
Poor Slight Fair Moderate Substantial Almost perfect

Kappa 0.0 .20 .40 .60 .80 1.0
Kappa Agreement Kappa Agreement

<0 Less than chance agreement 0.41 ~ 0.60 Moderate agreement
0.01 ~ 0.20 Slight agreement 0.61 ~ 0.80 Substantial agreement
0.21 ~ 0.40 Fair agreement 0.81 ~ 0.99 Almost perfect agreement
Table 3.3. Interpreting Kappa Coefficient3
Figure 4. Early detected carcinoma, <1cm in diameter, vertically orientated, halo sign(+), suspicious of neovascular
pattern, microcacification within mass. Both S-Detect and conventional US suggested malignancy.
Conlcusion
S-Detect is a promising tool in clinical application of breast imaging, based on BI-RADS1 classifications.
With substantial agreement to conventional method (k = 0.76), S-Detect contains clinical values in daily
practice. Moreover, with its provided accuracy (sensitivity 93.4%, specificity 62%, accuracy 79.3%),
S-Detect can certainly become a valuable tool in breast cancer screening, and even also assist beginners
and non-experienced radiologists.
Supported Systems References

- RS80A with Prestige 1. H
armien Zonderland and Robin Smithuis, “BI-RADS for Mammography and Ultrasound 2013”.
- RS80A RadiologyAssistant, Educational site of the Radiological Society of the Netherlands, 2013.
- HS70A 2. Nguyen Chan Hung , “Medicine of Tumors”, Vietnamese Medicine Publisher, 2004
3. A
nthony J. Viera and Joanne M. Garrett , “Understanding Interobserver Agreement: The Kappa
Statistic”, Family
ⓒ 2015 Samsung Medison All Rights Reserved.

Samsung Medison Reserves the right to modify any design, packaging, specifications
and features shown herein, without prior notice or obligation.
Please visit www.samsungmedison.com

White Paper S-Detect Breast 2 RS80A HS70A 2015

Uploaded by

Copyright:

Available Formats

White Paper S-Detect Breast 2 RS80A HS70A 2015

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

White Paper S-Detect Breast 2 RS80A HS70A 2015

Uploaded by

Copyright:

Available Formats

Article # WP WP201509-S-Detect / Issue Date 9 Sept, 2015

Evaluation of S-Detect as a screening

Results & Discussion

Malignant Benign Sum

2) Correlation between S-Detect and Conventional breast ultrasound

Poor Slight Fair Moderate Substantial Almost perfect

Kappa Agreement Kappa Agreement

Table 3.3. Interpreting Kappa Coefficient3

Supported Systems References

ⓒ 2015 Samsung Medison All Rights Reserved.

You might also like