Home About us Editorial board Ahead of print Current issue Archives Submit article Instructions Subscribe Search Contacts Login 
  • Users Online: 27
  • Home
  • Print this page
  • Email this page


 
 Table of Contents  
ORIGINAL ARTICLE
Year : 2020  |  Volume : 32  |  Issue : 2  |  Page : 140-144

CBCT-based active contour segmentation of bone invasion in oral squamous cell carcinoma - A preliminary retrospective study


Department of Oral Medicine and Radiology, Tamil Nadu Government Dental College and Hospital, Chennai, Tamil Nadu, India

Date of Submission16-Apr-2020
Date of Decision20-May-2020
Date of Acceptance26-May-2020
Date of Web Publication27-Jun-2020

Correspondence Address:
Dr. Shilpa Shree Kuduva Ramesh
Department of Oral Medicine and Radiology, Tamil Nadu Government Dental College and Hospital, No.1, Frazer Bridge Road, Muthuswamy Salai, Chennai - 600 003, Tamil Nadu
India
Login to access the Email id

Source of Support: None, Conflict of Interest: None


DOI: 10.4103/jiaomr.jiaomr_62_20

Rights and Permissions
   Abstract 


Context: Bone invasion by oral squamous cell carcinoma (OSSC) alone forms a predictor of overall prognosis, survival rate. Volume of actual bone osteolysis might give valuable information for treatment planning rather than planar measurements. Aim: To compare the manual and semiautomatic active contour segmentation in the volumetric analysis of mandibular bone invasion by OSCC through Cone Beam Computed Tomography (CBCT). Settings and Design: Hospital-based, preliminary, retrospective study. Methods and Material: Ten CBCT DICOM images of histologically confirmed cases of OSCC with frank bone invasion in the mandible were selected from the archive after satisfying our inclusion and exclusion criteria. Volumetrical analysis of tumor eroded area in the mandibular alveolus were done through ITK SNAP software (version 3.2) both by manual and semiautomatic segmentation and compared with each other. Statistical Analysis Used: R statistical computing software v3.6.3 (R core foundation, Vienna) was used. Intraclass correlation coefficient (ICC), Bland and Altman method, Passing Bablok analysis with Pearson's Correlation, Dice Similarity coefficient (DSC) were done for single examiner reliability, comparison of volumes and segmentation accuracy respectively. The third metric, that is, time needed for each segmentation method was also compared and statistically analysed. Results: Since Pearson's correlation coefficient of r = 1 and P value was 0.233 (>0.005), semiautomatic method of volumetric segmentation proved to be as accurate and reproducible as manual method without any significant volume difference. In addition to that, semiautomatic method was 10 times more rapid than manual method. Conclusions: This is the first kind of study in the literature showing the feasibility of active contour semiautomatic segmentation in the volumetric analysis of bone invasion caused by the OSCC which might be helpful in efficient oncological treatment planning.

Keywords: Cone beam computed tomography, invasion, squamous cell carcinoma, tumor, volumetric analysis


How to cite this article:
Kuduva Ramesh SS, Sadaksharam J. CBCT-based active contour segmentation of bone invasion in oral squamous cell carcinoma - A preliminary retrospective study. J Indian Acad Oral Med Radiol 2020;32:140-4

How to cite this URL:
Kuduva Ramesh SS, Sadaksharam J. CBCT-based active contour segmentation of bone invasion in oral squamous cell carcinoma - A preliminary retrospective study. J Indian Acad Oral Med Radiol [serial online] 2020 [cited 2020 Jul 11];32:140-4. Available from: http://www.jiaomr.in/text.asp?2020/32/2/140/288142




   Introduction Top


Oral squamous cell carcinoma (OSCC) is the most common malignant tumor that produce radiolucent lesions in the jawbones.[1] The AJCC staging system for OSCC classify tumors with bone invasion as T4 (stage IV) with resultant prognostic and management implications. Surgical resection of tumor, radiotherapy and its complications are based on the tumor localization and volume estimation.[2] At present, the radical surgical resection of all tissues infiltrated by the tumor with a safety margin of 7 mm remains the treatment of choice [3] which could be achieved when the volume of actual malignant invasion into the bone could be identified. Volumetric segmentations of bone invasion through Computed Tomography (CT) were done in previous studies but less radiation exposure, rapid acquisition time always make the CBCT as the best choice. Hence, hereby, we intend to determine the feasibility of semiautomatic active contour segmentation in the volumetric analyses of bone invasion by OSCC in CBCT images with a standard comparator of manual segmentation.


   Materials and Methods Top


After gaining permission from the department of oral medicine and radiology, ten CBCT DICOM files of histologically confirmed OSCC patients with known evidence of mandibular bony invasion, previously diagnosed in the orthopantomogram, were retrieved from the archive (Jan 2020 to March 2020). Images with artifacts that decreases diagnostic potential, pathologic fractures, other carcinomas were excluded. All scans were done using CS 9300 select CBCT machine (Carestream Health, Rochester, NY) with isotropic voxel size (slice thickness) of 180 m, 10 × 5 cm field of view with mandible as region of interest at 90 kVp, 4 mA, 8 s. Those DICOM files were transferred to an open source ITK-SNAP interactive software application (v3.2.0) (Penn Image Computing and Science Laboratory, University of Pennsylvania, USA) where both the manual and semiautomatic volumetric segmentation of bone invasion by OSCC were done and analyzed twice with interval of 2 weeks by a single maxillofacial radiologist.

Manual segmentation of the tumor invasion was performed by delineating its boundaries slice by slice in all three orientations, axial, coronal and sagittal using polygonal tool based on the gray value intensities [Figure 1].
Figure 1: Manual segmentation (a) Axial section (b) sagittal section (c) coronal section showing bone invaded area marked with dotted lines and white arrows

Click here to view


In semiautomatic segmentation, at first, region of interest was selected. Then “soft thresholding” mode was preferred where upper and lower threshold values were applied to a single image intensity component, that is, the erosion area. The difference between the probability that a voxel belongs to the object of interest and the probability that a voxel belongs to the image background is known as the speed image. Intensity values between the lower and upper thresholds were assigned positive speed values, and values outside the thresholds map to negative speed values. This method is called “Region Competition contour evolution.” A threshold between the soft tissue and bone gray value was determined. It was followed by geometric/geodesic active contour segmentation step, in which “snakes/seeds” were placed inside of the specific region of interest and connects in a way that adheres to the contour of the speed image with a geometric regularization term [Figure 2]. The segmentation results were visualized by a 3D mode of the segmented structures. The volume of the segmented structure (mm 3) was retrieved from the software, once the procedure had been finished [Figure 3].
Figure 2: Semiautomatic segmentation: Active Bubbles (seeds) (white arrows) placed in the region of interest (tumor eroded area) in the mandible after selecting the upper and lower threshold values (a) Axial (b) Sagittal (c) coronal

Click here to view
Figure 3: 3D reconstruction of bone invasion segment (a) Semiautomatic (b) Manual

Click here to view


Statistical analyses were done by R statistical computing software v3.6.3—”Holding the Windsock”—(R core foundation, Vienna). Intraclass correlation coefficient (ICC) for the quantitative evaluation of the single examiner reliability. Student “t” test was performed to compare volume difference of manual and semiautomatic segmentation. Significance of difference was set at P < 0.005. Bland and Altman's method was used to assess the degree of agreement between manual and semiautomatic segmentation methods. In this method, the difference between the measurements is plotted against their mean (which is considered to be the best estimate of the true values). Regression analysis was performed according to the Passing Bablok method. If the 95%confidence interval (CI) of the slope of the relationship (slope b) between two measurements that were being compared included 1, and the 95% CI of the ordinate of the relationship at the origin (intercept a) included 0, no statistically significant difference noted. Time needed for the manual and semiautomatic segmentations were compared. Dice similarity coefficient (DSC), a spatial overlap index of two binary segmentation images, was done by Convert 3D (c3d), a command line image processing tool that offers complementary features to ITK-SNAP, between the manual, automatic segmentations and among them. It ranges from 0, indicating no spatial overlap between two sets of segmentation to 1, indicating complete overlap.


   Results Top


ICC was 0.99 (95%CI) showing the reliability of the segmentations. Student “t” test showed P = 0.233 (>0.05) showing there is no significant volume difference between manual and semiautomatic segmentation methods [Table 1].
Table 1: Student “t” Test between manual and semi-automatic segmentations

Click here to view


Passing-Bablok regression analysis showed no significant volumetric difference between two methods with Pearson's Correlation coefficient was r = 1, Intercept of -0.00175, slope = 1 (95% CI), implying very high positive correlation between the volume obtained from the manual and semiautomatic segmentation [Figure 4]. Bland–Altman plot analysis showed mean difference of 0.0005 with upper 95% CI of 0.00219 and lower 95% CI of -0.0019 indicating that semiautomatic method is as accurate as manual measurement [Figure 5]. Dice similarity coefficient 0.96 ± 0.01 indicating almost complete overlapping between manual and semiautomatic segmentation. The mean ± SD time taken for manual segmentation was 31.3 ± 7.49 min, whereas semiautomatic segmentation took only 3.55 ± 0.95 min which is approximately 10 times more rapid than manual [Figure 6].
Figure 4: Passing Bablok regression fit plot showing no significant volume difference between two methods

Click here to view
Figure 5: Bland–Altman Analysis showing complete agreement between manual and semi-automatic segmentation. X-axis indicating the mean of manual and semiautomatic methods. Y axis indicating difference between manual and semiautomatic

Click here to view
Figure 6: Box Plot showing Time taken for Manual Segmentation and Semiautomatic segmentation

Click here to view



   Discussion Top


OSCC may invade into underlying bone from any direction, producing a polymorphous osteolytic lesion with ill-defined, ragged borders. Evidence of bone invasion around the teeth may first appear as widening of PDL space and loss of lamina dura. Floating teeth appearance is the striking feature of malignancy invading bone. As the stage advances, inferior border of the mandible may be thinned or destroyed and leads to pathologic fracture.[4] Bone invasion acts as a potential biomarker for the prognosis and overall survival rate. Medullary invasion alone was found to be an independent predictor of reduced overall prognosis and disease-specific survival of OSCC.[5] Two-dimensional measurements may misinterpret changes in bone invasion by OSCC as they neglect alterations in the third dimension.[6]

The accuracy for detection of mandibular bone invasion by OSCC, with sensitivity values are 94%, 95%, 83%, 55% by MRI, CBCT, CT, and panoramic radiography, respectively, and specificity values of 100% for CT, MRI, CBCT, 97% for PET/CT and 91.7% for panoramic radiography.[7] Recommendations for management decisions frequently rely on these imaging modalities. CBCT shows a much higher sensitivity for cortical bone invasion and a better negative predictive value.[8] With a significantly lower exposure dose, it can rule out the bone invasion effectively and prevent overtreatment. Considering the three-dimensional (3D) high-resolution images delivered by CBCT along with minimized artifacts in the mandible it provides an alternative imaging technique.[9]

Image segmentation can be seen as the biomedical image data set processing equivalent of surgical dissection. Image segmentation for volume characterization has shown to be promising and illustrates an unparalleled intrinsic repeatability and reliability in previous neuroimaging studies.[10],[11]

Volumetric analysis by image segmentation has been shown to be more accurate than unidimensional size characterization of tumors.[12] Elley et al. (2013) showed that tumor volume was significant at predicting all-cause survival and disease-free survival at 5 years in a 10-years MRI based retrospective study. Tumor volume stratification to correlate with the TNM staging system resulted in downstaging in 40 of the 62 cases.[13]

The appearance of teeth with similar intensity with the bone tissues may be a hindrance for automatic volumetric segmentation of bone invasion. Hence to separate the bone and tooth element from other elements, a threshold between the soft tissue and bone gray value was determined based on the study by Cuadros Linares et al. (2018).[14]

A discrepancy in the segmentation of bone invasion due to variable rates of periodontal disease in the study populations has been noticed in the previous studies as it was difficult to distinguish it from bone invasion and led to false-positive results. However, periodontal disease is usually associated with uniform bone loss on the buccal and lingual aspects of the teeth, whereas bone invasion is only seen on the side where the tumour is located. Additionally, it would be unusual, but not impossible, to have a single site of gross bone loss that just happened to be adjacent to a malignant tumour. From a clinical perspective, periodontal disease is easily differentiated from cancer by simple examination of the nature of the soft tissue and the lack of a “mass” in the former.[15]

In this retrospective study, we correlated the 3D segmentation of bone invasion/erosion in the mandible by squamous cell carcinoma between manual and semiautomatic methods. The positive correlation between manual and semiautomatic and significant statistical agreement, intrarater correlation showed the feasibility of semiautomatic segmentation of the bone invasion through CBCT images. This provides effective treatment planning with reduced radiation exposure.

Similar Volumetric analysis of Medication related osteonecrosis of jaw (MRONJ) lesions was studied via semiautomatic segmentation of CBCT images through ITK SNAP software by Zirk et al. (2019),[16] with region competition contour mode and there was no statistically significant difference between absolute and relative osteolysis volume and type and number of surgical interventions.

Safi et al. (2018) analyzed ameloblastoma with CBCT images volumetrically with the means of semiautomatic image segmentation and showed volumetric measurement (image segmentation based) and indicated statistically significant associations between volume and the radiological signs: cortical perforation, locularity, tooth resorption, local infection. When comparing the maximum ranges between the diameter and the volume, it has been noticed that the largest ameloblastoma was 53 times bigger than the smallest, whereas the maximum diameter was only 3.7 times larger than the smallest diameter actually concealing the immense size difference.[17] Similarly semiautomatic segmentation of KCOTs revealed their volumes were significantly higher the mean size of non-neoplastic odontogenic jaw cysts, adding an argument in favor of the neoplastic nature of KCOTs.[18]

There was no significant difference between manual segmentation and threshold-based segmentation in a similar study of segmentation of mandibular odontogenic tumors and cyst.[19] Another study compared the two methods of creating 3D replica of mandibular cysts and tumors using CT and CBCT DICOM images and showed no major differences between manual and automatic analysis. The use of the algorithm, however, has the advantage of rapidity which favors the use of semiautomatic segmentation method and also been proven in our study.[20]

A retrospective study correlated 3D-CT (3D computed tomography) volume measurements of malignant tumors with the response to treatment, and observed the bone invasion in these lesions, in which no false negatives were obtained using the 3D segmentation protocol.[21]

Our results are similar to dento-maxillary osteolytic image analysis by Valleys et al. (2018), comparing the reliability of a semiautomatic segmentation tool with manually defined segmentation in CBCT scans [22]

The future direction of our study is to include larger study sample with different region of interest, to correlate with the actual tumor volume prospectively and predicting sensitivity and specificity of semiautomatic segmentation of bone invasion via CBCT. Based on this semiautomatic segmentation study results, fully automated segmentation methods can be applied and can be correlated with other methods, eventually incorporation of artificial intelligence in the planning of radiotherapy and surgical resection based on the amount of bone osteolysis occurred.


   Conclusion Top


An accurate as well as rapid image segmentation modality in 3D with less radiation exposure parameters might be useful for the efficient cancer treatment planning and early diagnosis of the bone invasion. Based on our knowledge and literature search, this study becomes the first in showing the feasibility of region competition contour-based segmentation in the volumetric analysis of mandibular bone invasion by squamous cell carcinoma as a result of comparison with manual method.

Ethical considerations

Requisite ethical clearance was obtained from the Ethical Committee to carry out the study (Ref no OMR/2/2019TNGDCH).

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.



 
   References Top

1.
Wood NK, Goaz PW. Differential Diagnosis of Oral and Maxillofacial lesions. 5th ed. Missouri: Mosby; 1997. p. 360-2.  Back to cited text no. 1
    
2.
Nathu RM, Mancuso AA, Zhu TC, Mendenhall WM. The impact of primary tumor volume on local control for oropharyngeal squamous cell carcinoma treated with radiotherapy. Head Neck 2000;22:1-5.  Back to cited text no. 2
    
3.
Shah JP, Gil Z. Current concepts in management of oral cancer-surgery. Oral Oncol 2009;45:394-401.  Back to cited text no. 3
    
4.
Mallaya SM, Lam EWN. White and Pharaoh's Oral Radiology: Principles and Interpretation: South Asia. 2nd ed. Elsevier; 2019. pp. 494-99.  Back to cited text no. 4
    
5.
Ebrahimi A, Murali R, Gao K, Elliott MS, Clark JR. The prognostic and staging implications of bone invasion in oral squamous cell carcinoma. Cancer 2011;117:4460-7.  Back to cited text no. 5
    
6.
Greess H, Nömayr A, Tomandl B, Blank M, Lell M, Lenz M, et al. 2D and 3D visualisation of head and neck tumours from spiral-CT data. Eur J Radiol 2000;33:170-7.  Back to cited text no. 6
    
7.
Uribe S, Rojas LA, Rosas CF. Accuracy of imaging methods for detection of bone tissue invasion in patients with oral squamous cell carcinoma. Dentomaxillofac Radiol 2013;42:20120346.  Back to cited text no. 7
    
8.
Bombeccari GP, Candotto V, Giannì AB, Carinci F, Spadari F. Accuracy of the cone beam computed tomography in the detection of bone invasion in patients with oral cancer: A systematic review. Eurasian J Med 2019;51:298-306.  Back to cited text no. 8
    
9.
Hakim S, Wieker H, Trenkle T, Sieg P, Konitzer J, Holl-Ulrich K, et al. Imaging of mandible invasion by oral squamous cell carcinoma using computed tomography, cone-beam computed tomography and bone scintigraphy with SPECT. Clin Oral Invest 2013;18:961-7.  Back to cited text no. 9
    
10.
Yushkevich PA, Piven J, Hazlett HC, Smith RG, Ho S, Gee JC, et al. User-guided 3D active contour segmentation of anatomical structures: Significantly improved efficiency and reliability. Neuroimage 2006;31:1116-28.  Back to cited text no. 10
    
11.
Yushkevich PA, Yang Gao, Gerig G. ITK-SNAP: An interactive tool for semi-automatic segmentation of multi-modality biomedical images. Conf Proc IEEE Eng Med Biol Soc 2016;3342-5.  Back to cited text no. 11
    
12.
Dejaco D, Url C, Schartinger VH, Haug AK, Fischer N, Riedl D, et al. Approximation of head and neck cancer volumes in contrast enhanced CT. Cancer Imaging 2015;15:16.  Back to cited text no. 12
    
13.
Eley KA, Watt-Smith SR, Golding SJ. Magnetic resonance imaging–based tumor volume measurements predict outcome in patients with squamous cell carcinoma of the mandible. Oral Surg Oral Med Oral Pathol Oral Radiol 2013;115, 255-62.  Back to cited text no. 13
    
14.
Linares OC, Bianchi J, Raveli D, Neto JB, Hamann B. Mandible and skull segmentation in cone beam computed tomography using super-voxels and graph clustering. Vis Comput 2019;35;1461-74.  Back to cited text no. 14
    
15.
Czerwonka L, Bissada E, Goldstein DP, Wood RE, Lam EW, Yu E, et al. High-resolution cone-beam computed tomography for assessment of bone invasion in oral cancer: Comparison with conventional computed tomography. Head Neck 2017;39:2016-20.  Back to cited text no. 15
    
16.
Zirk M, Buller J, Zöller JE, Heneweer C, Kübler N, Lentzen M-P. Volumetric analysis of MRONJ lesions by semiautomatic segmentation of CBCT images. Oral Maxillofac Surg 2019;23:1-8.  Back to cited text no. 16
    
17.
Safi AF, Kauke M, Timmer M, Grandoch A, Nickenig HJ, Gultekin S, et al. Does volumetric measurement serve as an imaging biomarker for tumor aggressiveness of ameloblastomas? Oral Oncol 2018;78:16-24.  Back to cited text no. 17
    
18.
Kauke M, Safi AF, Grandoch A, Nickenig HJ, Zöller J, Kreppel M. Volumetric analysis of keratocystic odontogenic tumors and non-neoplastic jaw cysts – comparison and its clinical relevance. J Cranioaxillofac Surg 2017;46:257-63.  Back to cited text no. 18
    
19.
Rana M, Modrow D, Keuchel J, Chui C, Rana M, Wagner M, et al. Development and evaluation of an automatic tumor segmentation tool: A comparison between automatic, semi-automatic and manual segmentation of mandibular odontogenic cysts and tumors. J Craniomaxillofac Surg 2014;43:355-9.  Back to cited text no. 19
    
20.
Stoetzer M, Nickel F, Rana M, Lemound J, Wenzel D, von See C, et al. Advances in assessing the volume of odontogenic cysts and tumors in the mandible: A retrospective clinical trial. Head Face Med 2013;9:14.  Back to cited text no. 20
    
21.
Cavalcanti MG, Santos DT, Perrella A, Vannier MW. CT-based analysis of malignant tumor volume and localization. A preliminary study. Braz Oral Res 2004;18:338-44.  Back to cited text no. 21
    
22.
Vallaeys K, Kacem A, Legoux H, Le Tenier M, Hamitouche C, Arbab-Chirani R. 3D dento-maxillary osteolytic lesion and active contour segmentation pilot study in CBCT: Semi-automatic vs manual methods. Dentomaxillofac Radiol 2015;44:20150079.  Back to cited text no. 22
    


    Figures

  [Figure 1], [Figure 2], [Figure 3], [Figure 4], [Figure 5], [Figure 6]
 
 
    Tables

  [Table 1]



 

Top
 
 
  Search
 
Similar in PUBMED
   Search Pubmed for
   Search in Google Scholar for
 Related articles
Access Statistics
Email Alert *
Add to My List *
* Registration required (free)

   Abstract Introduction Materials and Me... Results Discussion Conclusion Article Figures Article Tables
  In this article
 References

 Article Access Statistics
    Viewed70    
    Printed0    
    Emailed0    
    PDF Downloaded28    
    Comments [Add]    

Recommend this journal


[TAG2]
[TAG3]
[TAG4]