the role of computer-aided detection and …...technical note the role of computer-aided detection...

+ MODEL

Journal of Medical Ultrasound (2015) xx, 1e8

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.jmu-onl ine.com

TECHNICAL NOTE

The Role of Computer-Aided Detection andDiagnosis System in the DifferentialDiagnosis of Thyroid Lesions inUltrasonography

Tien-Chun Chang*

Department of Internal Medicine, National Taiwan University Hospital and College of Medicine,National Taiwan University, Taipei, Taiwan

Received 7 July 2015; accepted 27 October 2015

KEYWORDScomputer-aideddetection anddiagnosis system,

fine-needle aspirationcytology,

thyroid cancer,thyroid nodule,thyroidultrasonography

Conflicts of interest: The author haby one the inventors of this method.* Correspondence to: Dr Tien-Chun

Road, Taipei, Taiwan.E-mail address: tienchunchang@nt

Please cite this article in press as: ChThyroid Lesions in Ultrasonography,

http://dx.doi.org/10.1016/j.jmu.20150929-6441/ª 2015, Elsevier Taiwan LLBY-NC-ND license (http://creativecom

Abstract To avoid abuse of fine-needle aspiration cytology (FNAC) and to save the time of alearning curve, a computer-aided detection and diagnosis (CAD) system to detect suspiciouslesions for FNAC from thyroid ultrasonography has been developed by the Department of Indus-trial Engineering, cooperative with the Department of Surgery at the National Taiwan Univer-sity, Taipei, Taiwan. The purposes of this article are to introduce how to utilize the CAD systemin thyroid ultrasonography, and to outline the real role of the CAD system. After marking theapparently transverse (extending across) axis and longitudinal axis of the nodule appearing byultrasonography, four parameters are calculated and displayed by the computer system auto-matically, which include microcalcifications, hypoechoic lesion, heterogeneity, and indistinctmargin. The results are displayed by the pointers in the semilunar figures. The necessity ofFNAC is dependent on the size and numbers of positive findings (pointers displayed in the rightside). This CAD system is objective and easy to use. It may supply an easy method to determinethe necessity for FNAC, but what we must keep in mind is that this method can reduce the ne-cessity of FNAC, not replace FNAC for the diagnosis of thyroid cancer.ª 2015, Elsevier Taiwan LLC and the Chinese Taipei Society of Ultrasound in Medicine. This isan open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

s never been involved in any funded projects developing the CAD device. All the figures are suppliedTherefore, readers must be careful to judge the feasibility of the method mentioned is this article.Chang, Department of Internal Medicine, National Taiwan University Hospital, 7, Chung-Shan South

u.edu.tw.

ang T-C, The Role of Computer-Aided Detection and Diagnosis System in the Differential Diagnosis ofJournal of Medical Ultrasound (2015), http://dx.doi.org/10.1016/j.jmu.2015.10.002

.10.002C and the Chinese Taipei Society of Ultrasound in Medicine. This is an open access article under the CCmons.org/licenses/by-nc-nd/4.0/).

http://creativecommons.org/licenses/by-nc-nd/4.�0/


mailto:[email protected]

http://dx.doi.org/10.1016/j.jmu.2015.10.002


www.sciencedirect.com/science/journal/09296441

http://www.jmu-online.com



2 T.-C. Chang

+ MODEL

Introduction

The number of cases of thyroid cancer have the trend ofincreasing gradually in many countries, including the UnitedStates [1] and Taiwan [2,3], although the increase in mor-tality is not noted. The dramatic increase in the number ofthyroid cancer cases is considered to be related to manyfactors, but increased screening by ultrasonography may bethe most important factor [1]. However, the increasedincidence across all tumor sizes suggests that increaseddiagnostic scrutiny is not the only reason [1].

Although it is cheap to carry out fine-needle aspirationcytology (FNAC) in Taiwan, especially under the coverage of

Figure 1 (A) Locate the first point in the left border of the thyroithe right border with the largest diameter of the lesion. Locate ththyroid nodule, then draw to the middle part in the lower border

Figure 2 For the longitudinal axis, locate the second point atthe middle part in the upper border of the thyroid nodule, thendraw to the middle part in the lower border of the thyroidnodule.

Please cite this article in press as: Chang T-C, The Role of Computer-AiThyroid Lesions in Ultrasonography, Journal of Medical Ultrasound (20

health insurance, the work-load would delay the furtherwork-up and management of patients if every case sus-pected to have thyroid nodule was to be aspirated and readafter staining to judge whether it is malignant or not.

In the study reported by Bonavita et al [4], there areseveral patterns shown by thyroid ultrasonography whichare indicative of benignancy, and further FNAC is notnecessary. These include spongiform configuration, cystswith colloid clots, giraffe pattern, and diffuse hyper-echogenicity. The specificity is 100%. They estimated 60% ofFNAC procedures could be saved. Different patterns onthyroid ultrasonograms between benign and malignant le-sions have also been stressed [5].

d lesion with the largest diameter. (B) Draw to extend across toe second point of the middle part in the upper border of theof the thyroid nodule.

Figure 3 As soon as the two axes are defined, the contour isautomatically generated by the computer-aided detection anddiagnosis system.

ded Detection and Diagnosis System in the Differential Diagnosis of15), http://dx.doi.org/10.1016/j.jmu.2015.10.002

Figure 4 Users can choose to fine-tune the contour by addingan additional marker, followed by clicking and dragging it.

Figure 5 The thyroid lesion is< 1 cm in diameter; fine-needle aspiration cytology needs to be performed.

Figure 6 The thyroid lesion from Figure 5. Parameter 1 (microcalcomputer-aided detection and diagnosis system are both positive.

CAD in Thyroid Lesion Differential Diagnosis 3

+ MODEL


Learning curves

It is not easy to read ultrasonography accurately for thebeginner, and the diagnosis by cytology is even more diffi-cult. Both of them need a so-called learning curve. Accu-mulation of experience can conquer this problem.However, at the initial stage, this problem can be solved bythe coverage of senior doctors. Although this is a formalway of learning medical knowledge, it is better to have afaster and objective way to solve this problem in the imagesystem.

Computer-aided detection or computer-aideddiagnosis

According to the recently published Food and Drug Admin-istration (FDA) guidance [6], computer-aided detection(CADe) devices are computerized systems intended toidentify, mark, highlight, or in any other manner directattention to portions of an image, or aspects of radiologydevice data, that may reveal specific abnormalities duringinterpretation of patient radiology images or patient radi-ology device data by the clinician, while the computer-aided diagnosis (CADx) devices include those that areintended to provide an assessment of disease or otherconditions in terms of the likelihood of the presence orabsence of disease, or are intended to specify disease type(i.e., specific diagnosis or differential diagnosis), severity,stage, or intervention recommended. The methodmentioned in this article therefore includes both CADe andCADx (CAD).

To differentiate benign from malignant thyroid lesionsautomatically in order to avoid the need of a learningcurve and to save working time, a review of CAD of thyroidnodule in thyroid ultrasonography was published [7]. Theauthors showed that the features can be categorized into(1) the sonographic features from the ultrasound images,and (2) the nonclinical features extracted from the ul-trasound images using statistical and data miningtechniques.

cification) and Parameter 2 (hypoechogenicity) detected by the


Figure 7 The thyroid lesion is< 1 cm in diameter; follow-uponly is required.

Figure 9 Thyroid lesions 1e1.5 cm in diameter, for whichfine-needle aspiration cytology needs to be performed.

4 T.-C. Chang

+ MODEL

CAD in Taiwan

With the cooperation of experts from the Department ofIndustrial Engineering and the Department of Surgery, Na-tional Taiwan University, Taipei, Taiwan, a CAD for thyroidnodule was developed successfully, and obtained the pat-ents from the FDA of the United States [8,9], Taiwan[10e12], China [13], Malaysia [14], and Singapore [15,16].They proved that the new CAD method is more sensitive andmore objective than traditional observation of ultrasonog-raphy in the detection and quantification of micro-calcifications [17]. This method is also correlated withconventional ultrasonic heterogeneity assessment, but canbe a better aid in the diagnosis of thyroid malignancy [18].

Figure 8 The thyroid lesion from Figure 7. Parameter 1 (microcalcomputer-aided detection and diagnosis system are both negative


How to perform CAD

It is simple to perform the CAD invented in Taiwan. The firstpoint in the left border of the thyroid lesion with the largestdiameter (Figure 1) is located, then drawn to extend acrossthe lesion to the right border with the largest diameter ofthe lesion. For the longitudinal axis, the second point at themiddle part in the upper border of the thyroid nodule islocated, and then drawn to the middle part in the lowerborder of the thyroid nodule (Figure 2). Then the margin isshaped automatically (Figure 3). If users are not satisfiedwith the initial contour generated automatically, theselected marker that the observer wants to move can beselected and dragged to a new location. Release of themouse button sets the contour shape. Therefore, users canfine-tune the contour by adding additional markers to the

cification) and Parameter 2 (hypoechogenicity) detected by the.


Figure 10 The thyroid lesion from Figure 9. Either Parameter 1 (microcalcification) or Parameter 2 (hypoechogenicity) detectedby the computer-aided detection and diagnosis system is positive.

Figure 11 Thyroid lesions 1e1.5 cm in diameter which needfollow-up only.

Figure 12 Thyroid lesion from Figure 11. Both Parameter 1 (microthe computer-aided detection and diagnosis system are negative.


+ MODEL


contour and moving their position (Figure 4). Then theseverity of the parameters are calculated and displayed bythe computer system automatically, which include micro-calcification, hypoechoic lesion, heterogeneity, and indis-tinct margin. The results are displayed by the pointers inthe semilunar figures. The necessity of FNAC is dependenton the size and number of positive findings (pointers dis-played on the right hand side) according to the guideline ofthe American Thyroid Association for differentiated thyroidcancer [19].

Figures 5 and 6 represent the example of thyroidlesions< 1 cm in diameter, for which FNAC needed to beperformed because Parameter 1 (microcalcification) andParameter 2 (hypoechogenicity) detected by the CAD wereboth positive.

Figures 7 and 8 represent the example of thyroidlesions< 1 cm in diameter which needed follow-up onlybecause Parameter 1 and Parameter 2 detected by the CADwere both negative.

calcification) and Parameter 2 (hypoechogenicity) detected by


Figure 13 The thyroid lesion is> 1.5 cm in diameter, whichneeds fine-needle aspiration cytology.

Figure 15 Thyroid lesions> 1.5 cm in diameter, which needfollow-up only.

6 T.-C. Chang

+ MODEL

Figures 9 and 10 represent the example of thyroid le-sions 1e1.5 cm in diameter, for which FNAC needed to beperformed because either Parameter 1 or Parameter 2detected by the CAD is positive.

Figures 11 and 12 represent the example of thyroid le-sions 1e1.5 cm in diameter which needed follow-up onlybecause both Parameter 1 and Parameter 2 detected by theCAD are negative.

Figures 13 and 14 represent the example of thyroidlesions> 1.5 cm in diameter which need FNAC because ofthe presence of any one of Parameters 1e4 found by theCAD.

Figures 15 and 16 represent the example of thyroidlesions> 1.5 cm in diameter, which need follow-up onlybecause Parameters 1e4 detected by the CAD are allnegative.

These examples show that for small lesions (< 1 cm indiameter), both Parameter 1 (microcalcification) andParameter 2 (hypoechogenicity) need to be positive, andthen FNAC is required. If there is a lesion with diameterfrom 1 cm to 1.5 cm, either microcalcification or hypo-echogenicity need to be positive. However, if the lesion

Figure 14 Thyroid lesion from Figure 13. The presence of anyheterogenicity, and blurring margin) detected by the computer-aid


detected by thyroid ultrasonography is> 1.5 cm in diam-eter, then any one of four parameters (microcalcification,hypoechoic lesion, heterogeneity, and indistinct margin)being positive is enough for the necessity of FNAC. There-fore, the necessity of FNAC depends on size and items ofpositive finding. Microcalcification and hypoechoic lesionare more important than heterogeneity and indistinctmargin in this system for small thyroid lesion (< 1 cm indiameter).

CAD cannot replace FNAC

This CAD system is objective and easy to use. It may supplyan easy method to determine the necessity for FNAC, butwhat we must keep in mind is that while this method canreduce the necessity of FNAC, it cannot replace FNAC in thediagnosis of thyroid cancer as many people believe.

one of Parameters 1e4 (microcalcification, hypoechogenicity,ed detection and diagnosis system.


Figure 16 Thyroid lesion from Figure 15. Parameters 1e4, detected by the computer-aided detection and diagnosis system, areall negative.


+ MODEL

When to perform FNAC and how to performFNAC at the time with CAD

Using the CAD system, the necessity for performing FNACcan be determined. If the lesion is not easily located, it isbetter under ultrasonography-guidance [20]. The method offurther management may be further planned with the helpof cytomorphology [21] or molecular determination [22].

Conclusion

CAD is an objective way to determine which thyroid nodulesneed to have FNAC. It saves workload, and there is nointerobserver variation. However, no examination is per-fect, which also includes CAD of the thyroid nodule.Therefore, follow-up is necessary if the thyroid lesions areconsidered not to require FNAC by CAD. It is important tofollow this rule to avoid the legal problems which may occurif CAD predicts the lesion does not require FNAC, but thelesion ultimately is malignant.

Acknowledgments

The figures used in this article are kindly supplied by Pro-fessor Argon Chen, Graduate Institute of Industrial Engi-neering, National Taiwan University.

References

[1] Chen AY, Jemal A, Ward EM. Increasing incidence of differ-entiated thyroid cancer in the United States, 1988e2005.Cancer 2009;115:3801e7.

[2] Chiou ST. 2014 Annual Report of Health Promotion Adminis-tration, Taipei, Taiwan. Health Promotion Administration,Ministry of Health and Welfare; 2014.

[3] Chiang CJ, Chen YC, Chen CJ, et al. Cancer trends in Taiwan.Jpn J Clin Oncol 2010;40:899e904.

[4] Bonavita JA, Mayo J, Babb J, et al. Pattern recognition ofbenign nodules at ultrasound of the thyroid: Which nod-ules can be left alone? AJR Am J Roentgenol 2009;193:207e13.


[5] Chang TC. Ultrasonic features of thyroid cancers and benignthyroid nodules for determining the necessity of fine needleaspiration cytology. J Med Ultrasound 2010;18:54e61.

[6] Food and Drug Administration. Guidance for Premarket Notifi-cation [510(k)] Submission for Computer Assisted Detection De-vices. United States Food and Drug Administration; July, 2012.

[7] Acharya UR, Swapna G, Sree SV, et al. A review on ultrasound-based thyroid cancer tissue characterization and automatedclassification. Technol Cancer Res Treat 2014;13:289e301.

[8] Chang KJ, Chen WH, Chen A, et al. Method for retrieving atumor contour of an image processing system. US patent USpatent 8,374,892. 12, Feb., 2013.

[9] Chang KJ, Chen WH, Chen A, et al. Quantification method ofthe feature of a tumor and an imaging method of the same. USpatent US 8,948,474. 3, Feb., 2015.

[10] Chang KJ, Chen WH, Chen A, et al. Method for retrieving atumor contour of an image processing system. Taiwan patentI428781. 1, Mar., 2014.

[11] Chang KJ, Chen WH, Chen A, et al. Method for retrieving atumor contour of an image processing system. Taiwan patentI474284. 21, Feb., 2015.

[12] Chang KJ, Chen WH, Chen A, et al. Non-invasive diagnosticmethod for thyroid tumor. Taiwanpatent I401064. 11, Jul., 2013.

[13] Chang KJ, Chen WH, Chen A, et al. Method for retrieving atumor contour of an image processing system. China patentCN 102156874 B. 29, Jul., 2015.

[14] Chang KJ, Chen WH, Chen A, et al. Method for retrieving atumor contour of an image processing system. Malaysia patentMY-152307-A. 15, Sep., 2014.

[15] Chang KJ, Chen WH, Chen A, et al. Method for retrieving atumor contour of an image processing system. Singaporepatent 173251. 31, Jul., 2013.

[16] Chang KJ, Chen WH, Chen A, et al. Quantification method ofthe feature of a tumor and an imaging method of the same.Singapore patent 192537. 26, May, 2015.

[17] Chen KY, Chen CN, Wu MH, et al. Computerized detection andquantification of microcalcifications in thyroid nodules. Ul-trasound Med Biol 2011;37:870e8.

[18] Chen KY, Chen CN, Wu MH, et al. Computerized quantificationof ultrasonic heterogeneity in thyroid nodules. UltrasoundMed Biol 2014;40:2581e9.

[19] American Thyroid Association (ATA) Guidelines Taskforce onThyroid Nodules and Differentiated Thyroid Cancer,Cooper DS, Doherty GM, et al. Revised American Thyroid As-sociation management guidelines for patients with thyroidnodules and differentiated thyroid cancer. Thyroid 2009;19:1167e214.


http://refhub.elsevier.com/S0929-6441(15)00095-8/sref1










































8 T.-C. Chang

+ MODEL

[20] Chang TC. The roles of ultrasonography and ultrasonography-guided fine-needle aspiration cytology in the planning of man-agement of thyroid cancers. J Med Ultrasound 2015;23:17e24.

[21] Shih SR, Chang YC, Li HY, et al. Preoperative prediction ofpapillary thyroid carcinoma prognosis with the assistance of


computerized morphometry of cytology samples obtained byfine-needle aspiration: preliminary report. Head Neck 2013;35:28e34.

[22] Syrenicz A, Koziołek M, Ciechanowicz A, et al. New insightsinto the diagnosis of nodular goiter. Thyroid Res 2014;7:6.















the role of computer-aided detection and …...technical note the role of computer-aided detection...

Documents