Three‐dimensional modelling as a novel interactive tool for preoperative planning for complex perianal fistulas in Crohn's disease

The aim of this study is to demonstrate the added value of three‐dimensional (3D) reconstruction models and artificial intelligence for preoperative planning in complex perianal Crohn's disease. MRI is the gold standard for diagnosis of complex perianal fistulas and abscess due to its high sensitivity, but it lacks high specificity values. This creates the need for better diagnostic models such as 3D image processing and reconstruction (3D‐IPR) with artificial intelligence (AI) algorithms.


INTRODUC TI ON
Perianal Crohn's disease (pCD) can present as complex fistulas in almost 80% of cases with recurrence rates of 19%-57% in some series [1]. These patients require three-dimensional (3D) ultrasound or magnetic resonance imaging (MRI) to recognize fistula tracts and occult abscess [2]. Nevertheless, even though MRI is the gold standard for diagnosis of complex perianal fistulas and abscess due to its high sensitivity, it lacks high specificity values (69% in some series) [3,4].
This creates the need for better diagnostic models such as 3D image processing and reconstruction (3D-IPR) with artificial intelligence (AI) algorithms. 3D-IPR models have demonstrated that they can be used as an important tool for the management of these complex cases as well as creating a more spatial and intuitive model for surgical strategy planning [5]. Therefore, they can provide useful information and adequate knowledge of the anatomical relationships of those anorectal fistulas with adjacent pelvic structures so suitable surgical management can be planned and performed [5,6]. Moreover, physical representation of fistula tracts improves the surgeon's understanding as well as patients' thoughts about surgical strategy [7].
It is important to emphasize that the main objective in locating abscess and treating fistula tracts is to control perianal sepsis and offer Crohn's disease (CD) patients the possibility to initiate otherwise contraindicated biological treatment.

ME THOD
This video-assisted study aims to describe an initial experience with 3D virtual modelling of patients' pelvic anatomy and its relationship with complex fistulas to provide a tool for surgical preoperative planning.
First, the AI algorithm was developed from three retrospective cases. This means that patients had a preoperative MRI and then surgery was performed. Then, a prospective study was performed to evaluate the utility of 3D-IPR obtained from MRI to guide surgical intervention. Postoperative MRI was performed to assess the resolution of abscess or fistulas as seen at 3D-IPR.
Three-dimensional IPR requires the use of MRI to construct a visual representation of the pelvis while AI uses an algorithm called a deep convolutional neural network to analyse and perform a diagnosis with visual imagery. Convolutional operation generates a 2D output, but using stacks of these algorithms/filters would generate a 3D output [8]. Moreover, 3D-IPR uses MRI preprocessing using 'bias field correction' algorithms and anisotropic diffusion filtering of images. The 3D surface was reconstructed by means of modified marching cube algorithms [9]. Each pelvic structure was assigned a specific colour; red was used for internal fistula orifices, light blue for fistula tracts and green for abscess. An online platform afforded quick access to the 3D model and its manipulation during surgical intervention.
In the four analysed cases the following variables were collected: if the internal fistula orifice(s) was or were located in the area indicated by the 3D reconstruction, if seton placement in the main fistula tracts was possible and if abscesses were adequately drained.
Postoperative MRI was performed to evaluate successful abscess drainage.
In these four cases the AI algorithm was successful, but findings need to be further validated in wider samples of patients.

COMPARISON WITH OTHER ME THODS
Endoanal ultrasound (EAUS), 3D-EAUS and MRI are still the ideal tools for the diagnosis of perianal disease and locating fistula tracts in pCD patients.
In some studies, the sensitivity and specificity of MRI in detecting fistulas were 87% and 69%, respectively. The specificity of EAUS can be as low as 43% [10]. Conversely, EAUS is superior to MRI at detecting internal fistula orifices, but specificity values for both modalities are considered to be diagnostically poor [10,11]. The accuracy rate of EAUS is 80%-89% for delimiting fistula tracts and 91% for detecting an internal fistula orifice [12]. Additionally, 3D-EAUS may be able to grade fistula complexity but MRI provides higher accuracy for evaluating secondary extensions [11].
Moreover, the accuracy of these imaging techniques does not guarantee a better surgical outcome. Some studies show conflicting results in the impact of preoperative EAUS and the outcome of perianal disease [10,12].
This translates into patients requiring multiple surgical procedures to treat abscess and fistulas due to high relapse rates, increased hospital stays and delayed optimal medical treatment due monoclonal antibody contraindication if the focus of perianal sepsis is not controlled.
The main possible advantage of using 3D-IPR in pCD patients is to reduce the number of surgical interventions necessary to control perianal sepsis and drain fistula tracts with setons to initiate adequate medical treatment. Furthermore, fewer surgical procedures may reduce the risk of anal sphincter damage and consequent anal incontinence. Another advantage is that 3D-IPR does not require additional MRI sequences, extra time during MRI or contrast to create the 3D reconstruction [7].
It is important to note that the software used to create 3D models during this research has the particularity that it uses mathematical AI algorithms to detect fistula tracts, perianal abscess and even recognize the internal fistula orifice. This is an advantage over other software packages that do not use this method and require manual confirmation [13]. Moreover, AI algorithms make it possible to detect anatomical structures that may sometimes be difficult to detect even by an expert radiologist. Given the required time to develop the 3D model, the use of the technology also depends on the possibility to delay surgery; however, under some circumstances it is possible to have the results within 1 day.
On the other hand, using 3D-IPR does not require dedicated training by radiologists as it is based on variables objectively obtained from a MRI scan.
The main disadvantages of this technique are its elevated costs, availability and time necessary to create the 3D model. Therefore, it is a tool that must be indicated in specific circumstances, such as those described above.

RE SULTS
Case 1 was a 14-year-old male patient with no other comorbidities presenting for the first time with a CD perianal abscess. The patient received initial treatment with azathioprine, which was changed to infliximab after resolution of the abscess. MRI was performed to provide the basis for a 3D-IPR model.
Preoperative axial MRI showed a primary intersphincteric fistula with a subcutaneous secondary tract. The inter-sphincteric fistula became trans-sphincteric at the anterior perineal raphe and produced a hemi-horseshoe abscess that extended all the way to the levator ani muscle. On the coronal segments a left supra-levator fluid collection was found following an intersphincteric route. All this information was then translated into the 3D-IPR.
During surgical intervention, the 3D model platform was continuously reviewed with a laptop. First, secondary fistula tracts were found and curettage performed. Then the left intersphincteric fistula was located and a seton placed with the help of the 3D model.
Moreover, as seen in the MRI and the 3D reconstruction, the intersphincteric route to the supra-elevator space was discovered as well as the anterior trans-sphincteric fistula described previously. A mushroom catheter (Petzer drain) was placed in the supra-levator space to allow further cleansing. Finally, setons were placed through subcutaneous secondary fistula tracts.
Postoperative MRI showed the catheter in the supra-levator space up to the prostate gland with the hemi-horseshoe abscess resolved. The mushroom catheter was removed after third dose of infliximab. The patient did not present any relapse of the abscess at a 6 month follow-up (Figure 1, Video S1).

CON CLUS IONS
Three-dimensional reconstruction of complex fistulas aids surgeons with an easier identification of secondary fistula, internal anal orifices or occult/deep abscess, which may otherwise be challenging to identify from MRI alone [5].
Three-dimensional model reconstruction does not offer additional information to that obtained with MRI, unless mathematical models are applied [9], but it improves conceptualization of complex fistulas as it provides a more realistic representation of the anatomy.  orifices and fistula tracts to reduce the number of surgical procedures required to treat these patients (see Table 1).

FU N D I N G I N FO R M ATI O N
None.

CO N FLI C T O F I NTER E S T S TATEM ENT
Dr García-Granero works for Cella Medical Solutions. He served as technical adviser for the current manuscript. All other authors have no conflict of interest to disclose.

AUTH O R CO NTR I B UTI O N S
SJMF Coinceived the article, collected and analysed data, and drafted the manuscript AGG coinceived the article, analysed data, and drafted the manuscript AOS collected and analysed data, and reviewed the article GP Coinceived the article and drafted the manuscript AOR collected and analysed data, and reviewed the article AGC collected and analysed data, and reviewed the article LB collected and analysed data, and reviewed the article DGV collected and analysed data, and reviewed the article MGC collected and analysed data, and reviewed the article FXGA analysed data, and reviewed the article for important intellectual content.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that supports the findings of this study are available in the supplementary material of this article.