The TNM classification for oral malignancies has been criticized for its upstaging to T4a when tumour involves styloglossus, hyoglossus, palatoglossus and genioglossus. The aims of this study were to (1) create an anatomical computer atlas of extrinsic tongue musculature, and (2) reassess the original staging of pre-treatment archived magnetic resonance images (MRI) of tongue carcinomas using the strict extrinsic muscle criteria. The anatomy of the extrinsic tongue muscles was mapped using images from the Visible Human Project (VHP) to create a computer model of the extrinsic tongue muscles. This was co-registered with 87 archived pre-staging MRI scans of tongue carcinomas to assess tumour ingress of the extrinsic tongue muscles. Of the 87 image sets reviewed, 16 were of superficial tumours not visible on MRI. In the remaining 71 cases that showed positive extrinsic muscle tumour ingress, 52% were upstaged from T1/2/3 tumours to cT4a based upon this finding. Extrinsic lateral and genioglossus muscle invasion did not predict occult cervical lymph node invasion or disease-related survival. In conclusion, tumour invasion of styloglossus or hyoglossus would result in the majority of lateral tongue tumours being staged T4a. Such stratification is of little clinical relevance, and an alternative more reliable method is required.
Squamous cell carcinomas of the oral cavity, like other tumours, are classified according to the International Union Against Cancer and the American Joint Committee on Cancer (AJCC) TNM staging criteria ( Table 1 ). Whilst widely used, this staging system has resulted in critical scrutiny as it does not accurately predict occult cervical nodal disease or survival. One of the most frequently criticized factors of the TNM system is automatic upstaging due to tumour ingress into named anatomical structures. For tongue tumours this includes the ‘deep’/extrinsic tongue muscles (genioglossus, hyoglossus, palatoglossus and styloglossus). However, these muscles lie in such proximity to the mucosal margin for the majority of their extent, that difficulty arises in anatomically distinguishing superficial from deep limits. The descriptor ‘deep’ does not accurately describe the complete morphology of these named muscles throughout their relationship within the oral cavity, indeed all of these muscles have both superficial and deep components. Accurate documentation of the anatomical paths of these extrinsic muscles is required to permit adherence to the TNM staging criteria. This is possible with magnetic resonance imaging (MRI), which provides an opportunity to confirm both normal anatomy and tumour extent, but tumour ingress may prevent confirmation of the accurate relationship to the extrinsic muscles.
|T1||Tumour 2 cm or less in greatest dimension.|
|T2||Tumour >2 cm but <4 cm in greatest dimension.|
|T3||Tumour >4 cm in greatest dimension.|
|T4a (oral cavity)||Tumour invades through cortical bone, into deep/extrinsic muscle of tongue (genioglossus, hyoglossus, palatoglossus and styloglossus), maxillary sinus, or skin of the face.|
The aims of this study were to: (1) create an anatomical computer model of the external tongue musculature to assist in extrinsic muscle identification in cases where the anatomy is distorted by tumour ingress, and (2) reassess the original staging of archived pre-treatment MRI oral tongue carcinomas using the model to assist in confirming the strict anatomical criteria.
Materials and methods
Creation of a VHF-based anatomic atlas of extrinsic tongue musculature
The anatomy of the extrinsic tongue muscles was mapped in three dimensions using images from the Visible Human Project (VHP). The VHP obtained high-resolution computed tomography (CT) and MRI images prior to freezing, sectioning, and photographing two human cadavers (male 1994; female 1995). This produced three-dimensional (3D) datasets of the complete human anatomy, available free to the scientific community. The Visible Human Female (VHF) is the more comprehensive of the datasets. Photographs are in 24-bit colour with a slice resolution of 2048 × 1216 pixels, and CT/MRI images have a pixel size of 0.33 mm × 0.33 mm and slice thickness of 0.33 mm (isometric voxels). The isometric parameters make the data ideal for multiplanar reformatting (MPR) and 3D rendering. The VHF provides unrivalled anatomical imaging of the lingual myostructure without the distortion effects and shrinkage of formalinized tissues, which can be navigated using an image stack similar to viewing sectional images of CT/MRI.
With the assistance of an anatomist, the boundaries of the extrinsic muscles were individually identified on each imaging slice of the VHF. Each muscle was manually segmented and sequential slices combined to produce a 3D computer model of their anatomical position. The model was validated and updated by comparison with existing models in a variety of forms, including point cloud, 3D renders of triangular meshes, and parallelepipeds. The accuracy of the model was confirmed by co-registration with MRI datasets from healthy volunteers.
Assessment of re-staging archived pre-treatment MRI oral tongue carcinomas using the strict anatomical criteria utilizing co-registration with the computer-generated anatomical model
Ninety-seven consecutive patients with oral tongue carcinomas who had undergone pre-treatment MRI prior to primary surgical ablation between 1999 and 2008 were identified from the departmental database and their pre-treatment MRI images reviewed. Of these, nine patients were excluded because the MRI images were of poor quality, and one case, because of image file corruption. As a result, 87 image sets were available for analysis. MRI images were viewed on eFilm (Version 1.83, Merge Healthcare, Chicago, IL, USA) and all available MRI pulse-sequences utilized to evaluate image quality for susceptibility and motion artefact. Patient demographic data were gathered from the departmental database in combination with a retrospective review of the patient records. Preoperative clinical tumour staging, according to the TNM classification, was completed at multidisciplinary review of the patient. This included clinical and radiological examination with a consensus opinion reached by the team.
The VHF-based anatomical computer model created in the first part of the study was co-registered with the pre-treatment MRI images ( Fig. 1 ; see also Supplementary Material, movie clip). The model was limited to the coronal plane for easy identification of the relevant anatomy, necessitating co-registration with T2-weighted coronal images for optimal tumour identification. The tumour-containing images were identified and exported as JPEG files. In order to identify the most appropriate model slice in which to morph the tumour-containing MRI image, a ratio was calculated of the distance from the tumour to the posterior limit of the tongue representing this distance as a proportion of the entire length of the tongue. This ratio was then multiplied by the total number of slices that were contained in the model ( n = 44), with the result rounded to the next whole slice to provide the most appropriate model slice.
The 3D model depicted in the video is that of the extrinsic muscles of the tongue: genioglossus (green), styloglossus (cyan), and hyoglossus (purple).
The MRI-derived JPEG and model slice were imported into FantaMorph (v4.1, Abrosoft, www.fantamorph.com ), a commercially available fusion software. FantaMorph uses a non-rigid image co-registration algorithm to warp the source image onto the recipient image. In this case the source image was the model slice and the recipient was the tumour-containing MRI slice. A maximum of 20 corresponding control points were manually defined on the source and recipient images to perform the non-rigid deformation ( Fig. 2 ). The majority of these points were placed on the tongue surface. The first four were placed at the cardinal directions, defining the superior, lateral, and inferior limits of the tongue contour. Where the midline of the tongue was visible, control points were added to compensate for deformation of the midline by any tumour mass effect. The resulting composite image was exported as a JPEG file and stored for reference. The presence of tumour invasion of the extrinsic muscle was assessed visually by two of the authors working in collaboration.
A 3D animation was subsequently created to aid anatomical visualization of the extrinsic muscles of the tongue, excluding palatoglossus (see Supplementary Material, movie clip). This was designed using references available from drawings, computer renderings, and an existing 3D model of a tongue and its extrinsic muscles based on the VHF project. The objects were constructed in a subdivision modeller (Wings3D, Open-source software) that permits derivation of organic models from two-dimensional references. The shape patterns were first reconstructed as rough polyhedra (with fewer than 50 vertices each) and then the polygons subdivided to obtain a smooth appearance.
Consent is routinely obtained from patients for the use of their medical records and imaging in audit and research. Ethical approval for this study, which had no direct impact on patient care, was not required in the UK.
Fig. 1 demonstrates the final computer model to assist in extrinsic tongue muscle identification.
The demographic data of the 87 patients included in the study are summarized in Table 2 . The mean age was 58 years (range 22–87 years). Of the 87 patients, 16 (18.4%) had superficial tumours not visible on MRI. These were designated as negative for extrinsic muscle involvement. Of the remaining 71 patients with positive extrinsic muscle tumour ingress, a total of 292 tumour-containing MRI slices were available for which the VHF–model fusion was completed. Based on visual inspection, 53 of the 87 cases (60.9%) were found to exhibit invasion of the lateral extrinsic tongue muscles. Nine of the 87 cases (10.3%) were found to have frank invasion of the genioglossus. Of the remaining 78 cases, 60 (76.9%) displayed no invasion of the genioglossus, while 19 (24.4%) were thought to reach but not invade the genioglossus.
|Clinical T class|
|Clinical N class|
|Clinical M class|
Utilizing these results, 37 of 71 (52.1%) cases originally staged cT1/T2/T3 were upstaged to cT4 ( Table 3 ).
Archived nodal and outcome data confirmed extrinsic muscle invasion did not predict occult lymph node metastasis. However, lateral muscle invasion did predict all-cause survival (ACS) and disease-free survival (DFS). Detailed results are shown in Tables 4–6 and Fig. 3 .