{"paper_id":"325226d4-4989-4f6e-ac46-4e9d74f46890","body_text":"1 \n \n \n \nUniversità di Pisa \nFacoltà di Medicina e Chirurgia \n \n                    Scuola di Specializzazione in Radiodiagnostica \nDirettore: Prof. Carlo Bartolozzi \n \n \n \n \n \nTesi di Specializzazione \n \n     Anatomical localization of deep infiltrating endometriosis:  \n3D MRI reconstructions \n \n \n \n \n \nRelatore: \nChiar.mo Prof. Carlo Bartolozzi \n \n \n \n                                                    Candidata: \n                                                                            Dr.ssa Federica Forasassi \n \n \n \n \n \n \n \nAnno Accademico 2009/2010 \n\n2 \n \nIndex \n \n \n1. Abstract         p. 3 \n2. Introduction         p. 5 \n3. Materials and methods:                                                                p. 7 \n \n                   3.1  Patient characteristics                                                 p. 7 \n \n                  3.2  MRI image acquisition protocol                                  p. 7 \n \n3.3 3D MRI reconstructions: the semi-automatic  \n     segmentation technique (ITK-SNAP software)            p. 8 \n                  3.4  Surgery Examination                                                  p. 13 \n \n                  3.5  Image Analysis                                                            p. 16 \n \n                  3.6 Methods of Analysis                                                     p. 16 \n \n               3.7  Statistical Analysis                                                       p. 18 \n \n4.  Results          p. 18  \n5.  Conclusions        p. 23  \n     6.  Bibliography        p. 26 \n \n \n \n \n\n3 \n \n1. Abstract \n \nPurpose \nThe goal of this study was to determine the accuracy of  3D MRI reconstructions obtained with \n \nsegmentation technique in the preoperative assessment of deep infiltrating endometriosis (DIE)  \n \nand in particular to evaluate rectosigmoid and bladder wall involvement. \n \nMaterials and methods \nInstitutional review board approval for this study was obtained, and each patient gave written \n \ninformed consent. \n \nFifty-seven consecutive patients with diagnosis of DIE who had undergone pelvic MRI at 1.5 T \n \nbefore surgery between 2007 and 2011, were retrospectively evaluated and 3D post-processed  \n \nin order to obtain a detailed mapping of DIE. A blinded reader interpreted images. \n \nMRI results were compared with surgical findings and were scored by using a four-point scale  \n \n(0_3 score). \n \n \n \n \nResults \n36/57 patients with symptomatic DIE underwent s urgery: 18/36 had endometriotic nodules \ninfiltrating the recto -uterine pouch,  12/36 the vescico -uterine pouch and 6/36 the rectovaginal \npouch.  \nThe sensitivity of MRI and 3D -MRI versus surgery was respectively 64% versus 83%; \ndiagnostic accuracy of  3D-MRI respect to MRI  alone was 86% versus 67% for localization;  \n86% versus 67% for dimension; 79% versus 58 % for rectosigmoid infiltration; 92 % versus 75% \nfor  bladder infiltration. \n \n\n4 \n \n \nConclusions \nIn this preliminary study 3D MRI reconstructions obtained with semi-automatic method of \nsegmentation provided encouraging results for staging DIE preoperatively. \nIn fact, the addition of 3D MRI reconstructions improved diagnostic accuracy and staging of \n \nDIE providing the exact volume of the lesions and enabling a precise mapping of these before \n \nsurgery. \n \n \nKey words:  Endometriosis; MRI; 3D reconstruction s; semi -automatic segmentation; wall \ninfiltration; laparoscopyc or robotic surgery. \n \n \n \n\n5 \n \n2. Introduction \n \n \nEndometriosis is defined as the presence of endometrial tissue outside the uterus; in particular,  \n \ndeep pelvic endometriosis, also called deep infiltrating endometriosis (DIE), is defined as  \n \ninfiltration of the implant of endometriosis under the surface of the peritoneum (5 mm in depth)  \n \n[1-4]. \n \nAlthough peritoneal endometriosis can be asymptomatic, DIE is a cause of pelvic pain,  \n \ndysmenorrhea, dyspareunia, dyschezia, and urinary symptoms and is frequently associated with \n \ninfertility. \n \nSince, recto-sigmoid involvement represents a severe form and occurs with a frequency of  \n \n6%–30% of cases of DIE, this disease can have a negative impact on everyday life and sexual  \n \nlife [5-7]. \n \nThe treatment of symptomatic DIE consists of complete excision of the lesions by means of \n \nlaparoscopy or robotic surgery.  \n \nTransvaginal sonography (TVUS) is recommended as first step for the diagnosis of  \n \nendometriomas and endometriosis of the bladder , but its value for the assessment of superficial  \n \nperitoneal lesions, ovarian foci, and DIE is poor. Moreover, sonography may not differentiate  \n \nsome endometriomas from hemorrhagic cysts or other ovarian neoplasms and is insensitive in  \n \nthe detection of peritoneal implants.  \n \nBecause of these limitations, laparoscopy or robotic surgery  have remained the standards of  \n \nreference for diagnosis and staging of pelvic endometriosis.  \n \nHowever, as laparoscopy and robotic surgery do not visualize well “atypical” non-pigmented  \n \nextraperitoneal sites of involvement and regions involved by pelvic adhesions, MR imaging  \n \n(MRI) has resulted the alternative and noninvasive technique for evaluation of endometriosis  \n \nshowing sensitivity and specificity of greater than 90% in the detection of endometriomas and  \n \nperitoneal implants [8-15]. \n \n\n6 \n \nTherefore, the diagnosis and staging of endometriosis should guide the surgeon to schedule the  \n \nmost appropriate one-step surgery: conservative laparoscopic/robotic surgery treatment or open \n \nsurgery with colonic resection if there is infiltrative parietal colon involvement [15-19]. \n \nThe addition of 3D MRI reconstructions, obtained with semi-automatic segmentation-technique,  \n \nimproved diagnostic accuracy and staging of DIE providing the exact volume of the lesions,  \n \nenabling a precise mapping of these before surgery and then planning the type of surgery  \n \ntreatment to be performed. \n \n \n \nThe goal of this study was to determine the accuracy of  3D MRI reconstructions obtained with \n \nsegmentation technique in the preoperative assessment of deep infiltrating endometriosis and in \n \nparticular to evaluate rectosigmoid and bladder wall involvement. \n \n \n\n7 \n \n3. Materials and methods \n3.1  Patient characteristics \n \nBetween  January 2007 and May  2011, 57 consecutive patients referred for pelvic MRI because \nof a clinical sus picion of endometriosis were prospectively enrolled. Among these, 36 patients \nwith symptomatic disease who underwent surgery (laparoscopic or robotic surgery) were \nincluded in our study (mean age, 28 years; range, 17–39 years).  \nEndometriosis was suspected because of one or more of the following symptoms: pelvic pain  \n \n(36 of 36  patients), dysmenorrhea, (19 of  36 patients), dyspareunia (25 of 36 patients),  \n \nrecurrence of symptoms and a past history of endometriosis (11 of 36 patients). \n \nExclusion criteria were the common contraindications to MRI (pacemaker, metallic foreign  \n \nbodies, and claustrophobia) and postmenopausal status. \n \n \n3.2  MRI image acquisition protocol \n \nMRI was performed with patients in the supine position by using a 1.5-T-whole-body MR  \n \nimager (Magnetom Symphony Maestro; Siemens Medical Solutions, Erlangen, Germany) with a  \n \npelvic phased-array coil. No contrast medium was used for imaging.  \n \nSequences acquired included thin-section high-spatial-resolution sagittal, axial and coronal  \n \nT2-weighted fast spin-echo images (FSE) with the following parameter: TR 4500 ms; TE107 ms;  \n \nslice thickness 3.0 mm; FoV 160 mm; Gap 0.8 mm; matrix 256 9 256; NEX 2. \n \nOur protocol included also T1-weighted FSE sequence with fat saturation (TR 500 ms, TE 15  \n \nms, slice thickness 3.0 mm, FoV 160 mm, Gap 0.8 mm, matrix 256 9 256, NEX 2) to detect  \n \nintra-nodal hematic signal and endometriomas. The total examination time was 15 min. \n \nRetrospectively, we post-processed MR studies previously performed, in order to obtain 3D MRI \n \nreconstructions. \n \n \n\n8 \n \n3.3  3D MRI reconstructions: the semi -automatic segmentation technique (ITK -SNAP \nsoftware) \n \nUsing MRI DICOM sagittal images radiologist made a semi-automatic segmentation of uterus,  \n \nvagina, sigma-rectum, bladder and endometriotic nodules by an open source software called  \n \nITK-SNAP (by Cognita Corporation) (Figure 1).  \n \n \n \n \n \n Figure 1. The layout of ITK-SNAP software \n \n \n \n \nImage segmentation plays a crucial role in many medical-imaging applications by  facilitating  \n \nthe delineation of anatomical structures and other regions of interest. \n \nSegmentation is the process of dividing images into constituent subregions where anatomical  \n \nstructures are indicated with different colors; the colors of the different organs were chosen by  \n \nthe operator to facilitate differentiation (Figure 2).  \n \n\n9 \n \n \n Figure 2. The colors of the different organs chosen by the operator to facilitate their differentiation \n \n \n \n \nSince the organs of the pelvic floor have complex three-dimensional (3D) structures, \n \nthree-dimensional virtual reality models of the female pelvic floor enhanced the anatomy of this  \n \ncomplex part of the body  [20-24].  \n \nA polygonal 3D model of each segmented structure was created by the semi-automatic program  \n \nwhich made also a surface smoothing (Figure 3a-b). \n \n\n10 \n \n \n Figure 3 a. Segmentation technique \n Figure 3 b. Segmentation technique \n \n \n \nMethods for segmentation of medical images are divided into three generations, where each  \n \ngeneration adds an additional level of algorithmic complexity.  \n \nThe first generation is composed of the simplest forms of image analysis such as the use of  \n \nintensity thresholds and region growing: this is  the earliest and lowest-level processing method. \n\n11 \n \n \nThe second generation is characterized by the application of uncertainty models and optimization  \n \nmethods, and the third generation incorporates knowledge into the segmentation process. \n \nThese generations indicate progress towards fully-automatic medical image segmentation and  \n \ntheir identification provides a framework for classifying the wide variety of methods that have  \n \nbeen devised   [25-28]. \n \nManual segmentation is possible but is a time-consuming task and subject to operator variability. \n \nReproducing a manual segmentation result is difficult and the level of confidence ascribed \n  \nsuffers accordingly. \n \nAutomatic methods are, therefore, preferable; however, significant problems must be overcome  \n \nto achieve segmentation by automatic means and it remains an active research area.  \n \nAutomatic segmentation methods are classified as either supervised or unsupervised. \n \nSupervised segmentation requires operator interaction throughout the segmentation process  \n \nwhereas unsupervised methods generally require operator involvement only after the  \n \nsegmentation is complete.  \n \nUnsupervised methods are preferred to ensure a reproducible result; however, operator  \n \ninteraction is still required for error correction in the event of an inadequate result  [29-32]. \n \nHowever, segmentation of medical images involves three main image related problems.  \n \nImages contain noise that can alter the intensity of a pixel such that its classification becomes  \n \nuncertain, images exhibit intensity nonuniformity where the intensity level of a single tissue  \n \nclass varies gradually over the extent of the image, and images have finite pixel size and are  \n \nsubject to partial volume averaging where individual pixel volumes contain a mixture of tissue  \n \nclasses so that the intensity of a pixel in the image may not be consistent with any one class. \n \nThese problems and the variability in tissue distribution among individuals in the human  \n \npopulation means that some degree of uncertainty must be attached to all segmentation results.  \n \nThis includes segmentations performed by medical experts where variability occurs between  \n \n\n12 \n \nexperts (inter-observer variability) as well as for a given expert performing the same  \n \nsegmentation on multiple occasions (intra-observer variability). \n \nDespite this variability, image interpretation by medical experts is generally considered to be the  \n \nonly available truth for in vivo imaging  [33]. \n \nAs far as concern the architecture of ITK-SNAP, it is based on “libraries” called Insight Toolkit  \n \n(ITK), Visualization Toolkit (VTK) and Fast LightToolkit (FLTK): they are all open source and  \n \ncross platform (Figure 4). \n \n \n   Figure 4.  3D reconstruction \n \n \nITK are libraries useful to implement algorithms of high level medical-image, from registration  \n \nand segmentation to filtering and quantitative measures; VTK are libraries for graphic  \n \nvisualization and image processing; FLTK are useful to create graphic interfaces for the users  \n \n(GUI). \n \nGraphic interface is very simple:  the default configuration is made by four windows \n \n(three for 2D and one for 3D navigation).  \n \nThe panel control has an intuitive graphic to which we can easily accede to all functionalities. \n\n13 \n \n \nWe focused primarily on the segmentation of magnetic resonance (MR) pelvis images although  \n \nmany methods can also be applied to other image types and to images from other modalities  \n \n[34]. \n \n \n3.4  Surgery Examination \n \n31/36 patients underwent laparoscopy and 5/36 patients underwent robotic surgery. \n \nFor each patient, independently from technique performed, the surgeon (more than 20 years of  \n \nexperience in the evaluation of endometriosis) gave us a detailed report that included the  \n \nlocation, the number of lesions and the their extension to pelvic organs, including the rectal and  \n \nthe bladder wall infiltration. \n \nIn 12/36 patients endometriotic nodules infiltrated bladder wall; in 19/36 patients infiltrated the  \n \nrectosigmoid serosa (13/36  rectouterine pouch; 6/36 rectovaginal pouch) and the surgeon  \n \nperformed meticulous dissection (“slicing or shaving”) to preserve the rectosigmoid wall \n \n(Figure 5 a-b-c; 6 a-b-c; 7 a-b; 9 a-b). \n              \n                                                 \n         \n                      \n             Figure 5 a-c. Non infiltrating recto-uterine pouch nodule; bladder wall infiltrating nodule      Figure 5 b \n \n\n14 \n \n                                                \n  \n                                                          Figure 5 c      \n                                                                   \n  \n                                                           \n  \n                                                      \n                                                  Figure 6 a-c. Infiltrating recto-uterine pouch nodule; \n                                                             vescico-uterine non infiltrating nodule \n \n        \n       \n  \n               Figure 6 b                                                                                                                            Figure 6 c \n \n\n15 \n \n \n                   \n  \n  Figure 7 a-c. Recto-vaginal pouch  non infiltrating nodule                                                              Figure 7 b   \n \n                         \n  \nFigure 8 a-b. Rectal wall infiltrating nodule                                                                                       Figure 8 b \n \n                      \n                 \n Figure 9 a-b. Non infiltrating recto-uterine nodule                                                                       Figure 9 b \n \n\n16 \n \n \n3.5  Image Analysis \n \nA  radiologist (with more than 8 years of experience in body MR imaging) who was blinded to  \n \nsurgical information,  analyzed all 3D MRI reconstructed images obtained with semi-automatic  \n \nsegmentation and completed a standardized form.  \n \nMRI, 3D MRI reconstructions and their correspondence with  surgery was graded on a scale  \n \nfrom 0 to 3 (0 _Inadequate, 1 _Poorly adequate,  2 _Fairly adequate, 3 _Perfectly coincident).  \n \nThe scale was based on the possibility of depicting the exact localization of the lesions  \n \n(the Douglas pouch, the vagina, the rectal wall, the vescico-uterine pouch and the bladder wall). \n \nDIE was described as nodular or retractile fibrotic-like tissue that was hypointense on  \n \nT2-weighted images and isointense to muscle on T1-weighted images. \n \nThe diagnosis of endometrioma was based on the identification of a cystic adnexal lesion with  \n \ncharacteristics of chronic bleeding: hyperintensity on fat suppressed T1-weighted images and a \n \nsignal intensity decrease (“shading”) onT2-weighted images.  \n \nAdhesions and indirect signs of adhesions were also described as hypointense peritoneal strands  \n \nthat converged to loculated fluid collections or organ displacements. \n \nIf rectal involvement was suspected, the precise location (distance from the anorectal junction) \n \nwas described and stated whether or not was present. Surgical reports were used as the reference  \n \nstandard. \n \n \n3.6 Methods of Analysis \n \n \na) Chi-square test in small numbers of observation: Fisher's exact test \n \nIn experimental practice it is frequently necessary to check whether there is agreement between  \n \nan observed distribution and the corresponding waited or theoretical distribution. \n \nThe test is defined “test for goodness of fit” for both qualitative data, that may be classified into \n \nnominal categories, both for quantitative data, distributed in classes of frequency: it is the  \n \n\n17 \n \npurpose for which it was proposed the chi square test. \n \nIt is one of nonparametric methods that is used to determine whether a series of data, collected in \n \nnature or in a laboratory, agreed with the specific hypothesis about their distribution or relative \n \nfrequency of classes. \n \nChi-square test is also used to compare two or more distribution observed. It is most frequently  \n \nused to verify the association between various modalities of two or more qualitative characters. \n \nIt is particularly useful in the initial phase of statistical analysis when it is necessary to look for  \n \nthe most significant variables and the relationship of association between them. \n \nIt may test a null hypothesis (0 H ), if the frequency distribution of certain events observed in a \n \nsample is consistent with a particular theoretical distribution, or to test an alternative hypothesis \n \nwhen it demonstrates the existence of  real difference even if the causes are unknown (1 H). \n \nThe choice between the two hypothesis is based on the estimated probability with the test. \n \nIt is the probability of finding by chance the observed distribution or a distribution that is further \n \naway from the expected in the condition that the null hypothesis is true. \n \nIf the probability calculated is small, the logic of statistical inference rejects the null hypothesis, \n \nimplicitly accepting the alternative hypothesis. \n \nPearson's chi-square is the best-known of several chi-square tests statistical procedures whose \n \nresults are evaluated by reference to the chi-square distribution. \n \nIts properties were first investigated by Karl Pearson in 1900. \n \nThe chi-square test is valid only for large samples. If the number of expected frequencies is small  \n \nin case of 2x2 tables must be used Fisher’s exact method. \n \n \nb) Cohen test \n \nK Cohen test is a measure of the agreement (coefficient of agreement) between the qualitative or \n \ncategorical responses given by two different people (inter-observer variation) or by the same  \n \nperson at different times (intra-observer variation) but  considering the same object. \n \n\n18 \n \nAgreement is considered to be good when was between 0.6 and 0.8 and was considered \n \nto be excellent when was greater than 0.8; a K<0.6 represent a disagreement index. \n \n \n3.7 Statistical Analysis \nThe data analysis evaluated the prevalence of infiltration detected by MRI and 3D MRI. \nSubsequently  it has been verified the sensitivity and specificity of  MRI versus 3D-MRI and \n \nthen the sensitivity of these two techniques compared to surgery (gold standard). \n \nBy means of analysis between the observed and expected values (chi-square test) has been \n \nevaluated the accuracy of MRI and 3D-MRI  to detect the exact site of the nodules of DIE, their \n \nvolume and the grade of parietal infiltration (rectum, bladder). \n \nIn particular, to underline the best accuracy of 3D-MRI in the evaluation of recosigmoid/bladder \n \ninfiltration, we estimated the intra-observer-agreement/disagreement (Cohen index). \n \nAgreement (k coefficient ) was considered excellent when it was > 0.8, good when it resulted \n \nbetween 0.6 and 0.8 and evaluated as a disagreement index  when <0.6. \n \n \n4 .Results \nIn 36 (100%) of 36 patients, DIE was confirmed at surgery and histopathologic examination. \n \n18/36 (50%) patients had endometriotic nodules infiltrating the recto-uterine pouch, 12/36(33%) \nthe vescico-uterine pouch and 6/36 (16%) the rectovaginal pouch.  \n10/36 patients (27%) had endometrioma correlat ing with DIE and another 4/36 (11%) had also \nadenomyosis.  \nMRI detected the infiltration of recto-sigma and bladder in 23/36 (63%) patients,  3D-MRI  \n \nincreased this percentage to  83% (30/36). In particular,  in the group of 18/36 patients with  \n \nrectouterine localization, MRI revealed the infiltration in  10/18 (56%) and 3D-MRI in 13/18   \n \n(72%); in the group of 12/36  patients with vescicouterine localization MRI revealed infiltration  \n \nin 9/12(75%) 3D-MRI in 11/12 (92%); in the group of 6/36 patients with rectovaginal  \n \n\n19 \n \nlocalization MRI revealed infiltration in 4/6 (67%) and 3D-MRI in 6/6 (100%) (Table 1). \n \n \n \n                    Table 1. Infiltration: MRI vs 3D-MRI \n \n \n \nIf 3D MRI is considered the gold standard method , the sensitivity of MRI amounted respectively \nto 77% in the recto -uterine pouch, 82% in the vescico -uterine pouch and 67% in the recto -\nvaginal pouch. These values pointed out the accuracy of 3D MRI to identify positive tests and its \nmajor sensitivity (Table 2). \n \n \n \n \n \n \n \n                             \n                       Table 2. MRI versus 3D-MRI \n \n \n \nMoreover were calculated the sensitivity of MRI and 3D-MRI versus surgery (Table 3).  \n \n \n \n            \n     \n \n \n                                                 Table 3. Sensitivity of MRI and 3D-MRI versus surgery \n \n \nAlso intra-observer agreement (analyzed with chi-square test) evidenced a difference statistically \n \n \n          Pts   MRI            % 3DMRI %  \nTotal 36  23/36 0.63 \n  \n30/36  0.83 \nRecto-uterine pouch 18  10/18 0.56  13/18 0.72 \nVescico-uterine pouch 12  9/12 0.75  11/12 0.92 \nRecto-vaginal pouch 4  4/6 0.67  6/6 1 \n \n     \n \n MRI vs 3D-MRI       \n  Sensitivity (%) Specificity (%) PPV (%) NPV (%) \nRecto-uterine pouch      77 (10/13)       100 (5/5)    100 (10/10)    62 (5/8) \n \nVescico-uterine pouch      82 (9/11)       100 (1/1)    100 (9/9)    33 (1/3) \nRecto-vaginal pouch      67 (4/6)              -    100 (4/4)          - \n  \n       MRI vs              \n      surgery \n        3D-MRI vs \n          surgery \nSensitivity (%) 64% 83% \n\n20 \n \n significative (p< 0.05)  between the two methods (Table 4). \n \n \n \n                                            \n                                                          \n \n \n \n                                           Table 4. Intra-observer agreement \n \nThe analysis of qualitative data , useful to identify the size of nodules, their localization  and the \ninfiltration of contiguous structures supported the best accuracy  of 3D -MRI respect to MRI \nalone.   \nIn particular as far as concerned localization 3D-MRI resulted perfectly coincident (86% ) \nrespect to MRI  alone (67%) (p<0.000), (Graph 1);  regard to dimension 3D-MRI resulted \nperfectly coincident (86% ) respect to MRI  alone (67%) (p<0.000), (Graph 2); the rectosigmoid \ninfiltration resulted perfectly coincident with 3D-MRI (79%) versus MRI alone (58%).  \nThe poorly coincidence was 42% for MRI and 21% for 3D-MRI (p<0.006), (Graph 3); the \nbladder infiltration resulted perfectly coincident with 3D-MRI (92%) versus MRI alone (75%). \nThe poorly coincidence was 25% for MRI and 8% for 3D-MRI (p<0.003), (Graph 4). \n \n \n             Intra-observer agreement between  \n                           MRI and 3D-MRI \n  P \nLocalization 0,000 \nDimension 0,000 \nBladder infiltration  0,003 \nRectum infiltration 0,006 \n\n21 \n \n \n \n \nGraph 1.  3D vs MRI: localization \n \n \n \n \n \n \n \n \n \n \nGraph 2.  3D vs MRI: dimension \n \n \n \n \n \n \n \n0 \n5 \n10 \n15 \n20 \n25 \n30 \n35 \nInadequate Poorly \nadequate \nFairly \nadequate \nPerfectly \ncoincident \n3D MRI \n0 \n5 \n10 \n15 \n20 \n25 \n30 \n35 \nInadequate Poorly \nadequate \n \nFairly \nadequate \n  \nPerfectly \ncoincident \n \n3D MRI \n\n22 \n \n \nGraph 3.  3D vs MRI: rectal infiltration \n \n \n \n \nGraph 4.   3D vs MRI: bladder infiltration \n \n \nIn order to  confirm the difference s between  MRI and 3DMRI findings detected by means  of  \nradiologist qualitative analysis , it  was c alculated Cohen’s K coefficient: f or each evaluated \nfinding (localization, dimension, parietal infiltration) the K Cohen indexes resulted lower than \nconcordance parameters, demonstrating the effective discordance  (all K coefficients were < 0.6) \nbetween MRI and 3DMRI and than validating the greater efficacy of MRI supported by 3D \nreconstructions. \n \n \n \n \n0 \n2 \n4 \n6 \n8 \n10 \n12 \n14 \n16 \n18 \n20 \nInadequate \n \nPoorly \nadequate \n \n \nPerfectly \ncoincident \n \n3D MRI \nFairly \nadequate \n  \n0 \n2 \n4 \n6 \n8 \n10 \n12 \nInadequate \n \nPoorly \nadequate \n \nFairly \nadequate \n \nPerfectly \ncoincident \n 3D MRI \n\n23 \n \n \n \n \n                \n               \n \n \n               \n \n \n                        \n                       Table 5.  Agreement/disagreement obtained analyzing MRI and 3DMRI findings \n \n \n5. Conclusions \n \nOur study results showed an excellent correlation of 3D MRI reconstructions with surgical  \n \nexamination findings of DIE. In particular, 3D MRI reconstructions detected the specific  \n \nlocations and dimension of the endometriosis nodules with a higher sensitivity (83%) and  \n \ndiagnostic accuracy (86%).  \n \nDue to this, the addiction of 3D MRI reconstruction in the future could influence the decision of  \n \nthe surgeon to perform laparoscopy or robotic surgery rather than more extensive surgery, \n \nproviding the precise preoperative mapping of deep endometrial lesions especially in forms of  \n \nadvanced endometriosis (stage IV) infiltrating rectosigmoid colon or bladder. \n \nIn our series, the prevalence of rectosigmoid wall involvement was high and the correlation  \n \nbetween 3D MRI and surgery just about the degree of invasion of this structure resulted excellent  \n \n(Cohen k coefficient of  0.28). \n \nOn the hand anterior deep endometriosis is much less frequent than posterior involvement and  \n \ninvolves the vescicouterine pouch and the bladder. Bladder involvement occurs in 2-6.4 % of  \n \npatients  [6]  and also at this level 3D MRI imaging resulted accurate (Cohen k coefficient of   \n \n0.39). \n \nIn our study, we had 12/36 cases of anterior deep endometriosis that infiltrated the bladder: while \n \n                         Concordance of MRI and 3DMRI findings   \n  K di Cohen \nLocalization 0,16 \nDimension \n \n0,18 \nBladder infiltration \n \n0,39 \n \nRectal infiltration 0,28 \n \n\n24 \n \nconventional MRI correctly identified 8/12 cases, 3D MRI reconstructions depicted correctly  \n \n11/12  cases of bladder involvement identifying also the smaller lesion (<1 cm): we missed one  \n \ncase because the bladder was empty. \n \n18/36 patients had the endometriosis nodules localized into the rectouterine pouch: 10/18 cases  \n \nwere correctly identified at conventional MRI, while 3D MRI reconstructions depicted correctly  \n \n13/18 cases of rectosigmoid wall involvement and any case was missed. \n \n6/36 patients had the nodules localization into the rectovaginal pouch: 4/6 cases were correctly  \n \nidentified at conventional MRI, while 3D MRI reconstructions we depicted correctly 6/6 cases of  \n \nrectal wall involvement and any case was missed. \n \nThe assessment of endometriosis may be laborious for the patient, who may undergo multiple  \n \nexaminations like transvaginal US, transrectal US, barium enema, cystoscopy, and rectoscopy.  \n \nInstead, MRI imaging supported by 3D MRI reconstructions enabled complete visualization of  \n \nthe pelvis and therefore could be the preoperative imaging technique of choice for the \n \nassessment of patients with a clinical suspicion of DIE also assessing the infiltration of   \n \nrectosigmoid and bladder walls.  \n \nFuture research in the segmentation of medical images will strive toward improving the  \n \naccuracy, precision, and computational speed of segmentation methods, in order to reduce the  \n \namount of  manual interaction by incorporating prior information  from atlases and by combining  \n \ndiscrete and continuous spatial-domain segmentation methods [20]. \n \nNot all 57 patients underwent surgery, so we do not know the accuracy of 3D MRI  \n \nreconstructions in our total study population; including only patients who underwent surgery  \n \nlikely skewed our results, because of the poorly population study.  \n \nOur surgeon knew the results of the MRI examination,  and also this could have biased results at  \n \nsurgery.  \n \nFinally, because our population was selected in a referral center, it was biased for the prevalence  \n \nof endometriosis. \n\n25 \n \n \nIn conclusion, pelvic MRI supported by 3D MRI reconstructions provided encouraging results  \n \nfor the diagnosis and the preoperative planning of DIE providing the exact volume of the lesions  \n \nand enabling a precise mapping of these before surgery  resulting particularly valuable as image- \n \nguided surgery techinque, in wich visualization of anatomy is a critical component.  \n \n \n \n \n\n26 \n \n 6.  Bibliography \n \n1. K.Kinkel, K.A.Frei, C.Balleyguier, C.Chapron. Diagnosis of endometriosis with imaging: \n     a review. Eur Radiol (2006) 16: 285–298DOI 10.1007/s00330-005-2882-y \n2. Cornillie FJ, Oosterlynck D, Lauweryns JM,Koninckx PR. Deeply infiltrating pelvic \n \n          endometriosis: histology and clinical significance. Fertil Steril 1990;53(6):978–983. \n \n3. Koninckx PR, Meuleman C, Demeyere S, Lesaffre E, Cornillie FJ. Suggestive evidence \n \n            that pelvic endometriosis is a progressive disease, whereas deeply infiltrating endometriosis \n \n           is associated with pelvic pain. Fertil Steril 1991;55(4):759–765. \n \n4. Anaf V, Simon P, El Nakadi I, et al. Relationship between endometriotic foci and nerves \n \n            in rectovaginal endometriotic nodules. Hum Reprod 2000;15(8):1744–1750. \n \n5. Anaf V, Simon P, El Nakadi I, Simonart T, Noel J, Buxant F. Impact of surgical resection \n \n            of rectovaginal pouch of Douglas endometriotic nodules on pelvic pain and some elements  \n \n            of patient’s sex life. J Am Assoc Gynecol Laparosc 2001;8(1):55–60. \n \n6. Chapron C, Fauconnier A, Vieira M, et al. Anatomical distribution of deeply infiltrating \n \n            endometriosis: surgical implications and proposition for a classification. Hum Reprod \n \n           2003;18(1):157–161. \n \n7. Chapron C, Chopin N, Borghese B, et al. Deeply infiltrating endometriosis: pathogenetic \n \n            implications of the anatomical distribution. Hum Reprod 2006;21(7):1839–1845. \n \n8. Nisolle M, Donnez J. Peritoneal endometriosis, ovarian endometriosis and adenomyotic \n \n            nodules of the rectovaginal septum are three different entities. Fertil Steril 1997;68(4): \n \n           585–596. \n \n9.  Re´genet N, Me´tairie S, Cousin GM, Lehur PA. Colorectal endometriosis: diagnosis and \n \n             management [in French]. Ann Chir 2001; 126(8):734–742. \n \n      10 . Chopin N, Vieira M, Borghese B, et al. Operative management of deeply infiltrating \n \n\n27 \n \n endometriosis: results on pelvic pain symptoms according to surgical classification.  \n \n       J. Minim  Invasive Gynecol 2005;12(2):106–112. \n \n11.  Bazot M, Darai E, Hourani R, et al. Deep pelvic endometriosis: MR imaging for  \n \n                   diagnosis and prediction of extension of disease. Radiology 2004;232(2):379–389. \n \n12.  Kinkel K, Frei KA, Balleyguier C, Chapron C.Diagnosis of endometriosis with \nimaging: \n \n       a review. Eur Radiol 2006;16(2):285–2987. \n \n13. Reinhold C, McCarthy S, Bret PM, et al. Diffuse adenomyosis: comparison of  \n \nEndovaginal US and MR imaging with histopathologic correlation.  \n \nRadiology 1996;199(1): 151–158. \n \n14 . Togashi K, Nishimura K, Kimura I, et al. Endometrial cysts: diagnosis with MR  \n \n      imaging. Radiology 1991;180(1):73–78. \n \n            15.  Woodward PJ, Sohaey R, Mezzetti TP Jr. Endometriosis: radiologic-pathologic \n \n       correlation. RadioGraphics 2001;21(1):193–216. \n \n16.  Kinkel K, Chapron C, Balleyguier C, Fritel X, Dubuisson JB, Moreau JF. Magnetic \n \n       Resonance imaging characteristics of deep endometriosis. Hum Reprod  \n \n      1999;14(4):1080–1086. \n \n 17.  Del Frate C, Girometti R, Pittino M, Del Frate G, Bazzocchi M, Zuiani C. Deep  \n \n         Retroperitoneal pelvic endometriosis: MR imaging appearance with laparoscopic  \n \n        correlation. RadioGraphics 2006;26(6):1705–1718. \n \n 18.  Vercellini P, Meschia M, De Giorgi O, Panazza S, Cortesi I, Crosignani PG. Bladder \n \n        detrusor endometriosis: clinical and pathogenic implications. J Urol 1996; \n \n       155(1):84–86. \n \n19.  Balleyguier C, Chapron C, Dubuisson JB, et al. Comparison of magnetic resonance  \n \n        Imaging and transvaginal ultrasonography in diagnosing bladder endometriosis.  \n \n        J Am Assoc Gynecol Laparosc 2002;9(1):15–23. \n \n\n28 \n \n20.  Annu. Rev. Biomed. Eng. 2000. 02:315–37 Currentmethods in medical image \n \n      Segmentation. Dzung L. Pham, Chenyang Xu2, and Jerry L. Princ.e \n \n      Department of Electrical and Computer Engineering, The Johns Hopkins University, \n \n      Baltimore, Maryland. \n \n      21. Grimson WEL, Ettinger GJ, Kapur T, Leventon ME, Wells WM, et al. 1997. Utilizing \n \n      segmented MRI data in image-guided surgery. Int. J. Pattern Recognit. Artif. Intell. \n \n      11:1367–97 \n \n22.  Haralick RM, Shapiro LG. 1985. Image segmentation techniques. Comput. Vis. \n \n       Graph. Image Proc. 29:100–32 \n \n23.  Gonzalez RC, Woods RE. 1992. Digital Image Processing. Reading, MA:  \n \n      Addison-Wesley. 716 pp. \n \n24.  Pal NR, Pal SK. 1993. A review on image segmentation techniques. Pattern Recognit. \n \n       26:1277–94 \n \n25 . Langan DA, Modestino JW, Zhang J. 1998. Cluster validation for unsupervised \n \n       stochastic model-based image segmentation. IEEE Trans. Image Process.  \n \n       7:180–95 \n \n26 .  Khaneja N, Miller MI, Grenander U.1998. Dynamic programming generation \n \n       of curves on brain surfaces. IEEE Trans. Pattern Anal. Mach. Intell.  \n       \n       20:1260–65 \n \n27.  Pham DL, Prince JL. 1999. An adaptive fuzzy c-means algorithm for image \n  \n       segmentation in the presence of intensity inhomogeneities. Pattern Recognit. Lett. \n  \n      20:57–68 \n \n28.  D.L. Pham, C. Xu, J.L. Prince, “Current methods in medical image \n \n       segmentation,” Annu. Rev. Biomed. Eng., vol. 2, pp. 315-337, 2000. \n \n29.  Medical Image Segmentation: Methods and Software. \n       \n       D.J. Withey and Z.J. Koles1, Department of Electrical and Computer Engineering,  \n\n29 \n \n \n       Department of Biomedical Engineering. University of Alberta, Edmonton, Alberta,  \n \n      Canada. \n \n30.  L.P. Clarke, R.P. Velthuizen, M.A. Camacho, J.J. Heine, M. Vaidyanathan, \n \n       L.O. Hall, R.W. Thatcher, M.L. Silbiger, “MRI segmentation: Methods and \n \n       applications,” Magn. Reson. Imaging, vol. 13, no. 3, pp. 343-368, 1995. \n  \n 31.  J.C. Bezdek, L.O. Hall, L.P. Clarke, “Review of MR image segmentation \n \n        techniques using pattern recognition,” Med. Phys., vol. 20, no. 4, pp. 1033- \n \n       1048, 1993. \n \n32.  S.D. Olabarriaga, A.W.M. Smeulders, “Interaction in the segmentation of \n \n       medical images: A survey,” Med. Image Anal., vol. 5, pp. 127-142, 2001. \n \n34.  Facoltà Di Ingegneria-CDLS in Ingegneria Biomedica. \n \n       Segmentazione di Immagini MDCT per la Pianificazione di Interventi di \n \n       Chirurgia Pancreatica tramite Visualizzazione 3D. Prof.Andrea Pietrabissa,  \n \n       Ing. Vincenzo Ferrari, Ing.Marina Carbone.","source_license":"CC0","license_restricted":false}