shape modeling of the paranasal sinuses to estimate natural - PowerPoint PPT Presentation

Automatic segmentation and statistical shape modeling of the paranasal sinuses to estimate natural variations Ayushi Sinha a , Simon Leonard a , Austin Reiter a , Masaru Ishii b , Russell H. Taylor a and Gregory D. Hager a a Dept. of Computer Science, The Johns Hopkins University, Baltimore, MD 21218, USA b Dept. of Otolaryngology-Head and Neck Surgery, Johns Hopkins Medical Institutions, Baltimore, MD 21287, USA

Introduction • Functional endoscopic sinus surgery (FESS) is a routine operation performed by an otolaryngologist • Between 200,000 and 600,000 endoscopic interventions per year in the USA [1][2][3] • Navigation during surgery can be improved using pre-operative CT • Reduces likelihood of potential complications • Enhances patient safety and outcome [1] Hosemann W, Draf C. Danger points, complications and medico-legal aspects in endoscopic sinus surgery. GMS Current Topics in Otorhinolaryngology, Head and Neck Surgery. 2013;12:Doc06. [2] Hepworth EJ, Bucknor M, Patel A, Vaughan WC. Nationwide survey on the use of image-guided functional endoscopic sinus surgery. Otolaryngol Head Neck Surg. 2006 Jul;135(1):68 – 73. [3] Psaltis AJ, Soler ZM, Nguyen SA, Schlosser RJ. Changing trends in sinus and septal surgery, 2007 to 2009. Int Forum Allergy Rhinol. 2012 Sep-Oct;2(5):357 – 361.

Nasal Cycle • Alternating partial congestion and decongestion of the nasal cavities due to the expansion and contraction of the inferior, middle, and superior turbinates [4] • Each cycle can span between ~50 minutes to several hours [5] • Need to compensate for this regularly deforming topology [4] Hasegawa M, Kern EB, The human nasal cycle. Mayo Clinic Proceedings. May 1977;51:28-34 [5] Atanasov AT. Length of Periods in the Nasal Cycle during 24-Hours Registration. Open Journal of Biophysics. 2014;4:93-96

Can we estimate this deformation? • Lack of longitudinal studies • But, plenty of head CTs from different patients • Can we characterize this deformation from head CTs of a large population?

Can we estimate this deformation? • Hypothesis is: Given CTs of 𝑜 individuals, it is likely that the turbinates of each individual are at a different state in the nasal cycle than all others. Therefore, a statistical model of the turbinates built from these 𝑜 CTs should also reflect natural variation.

Method Patient CTs Statistical Shape Model (SSM) Deformably Register Template Deformed Template Deform Template PCA

Template Creation [6] 1. Automatic Segmentation Template Creation Deformable Registration Segmentation Improvement 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [6] BB Avants, P Yushkevich, J Pluta, D Minko, M Korczykowski, J Detre, JC Gee, “The optimal template effect in hippocampus studies of diseased populations," NeuroImage 49(3), p. 2457, 2010.

Automatic Segmentation [7] Deformation Fields Deformed Meshes 1. Automatic Segmentation Template Creation Template Mesh Deformable Registration Segmentation Improvement 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [7] BB Avants, NJ Tustison, . Song, PA Cook, A Klein, and JC Gee, “A reproducible evaluation of ANTs similarity metric performance in brain image registration," NeuroImage 54(3), pp. 2033-2044, 2011.

Deformable Registration (DR) [7] 1. Automatic Segmentation Template Creation Deformable Registration Segmentation Improvement 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [7] BB Avants, NJ Tustison, . Song, PA Cook, A Klein, and JC Gee, “A reproducible evaluation of ANTs similarity metric performance in brain image registration," NeuroImage 54(3), pp. 2033-2044, 2011.

Gradient Vector Flow (GVF) [10][11] 1. Automatic Segmentation Template Creation Deformable Registration Segmentation Improvement 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [10] C. Xu and J. L. Prince, “Gradient vector ow: A new external force for snakes," in Computer Vision and Pattern Recognition, IEEE Computer Society Conference on, pp. 66-71, 1997. [11] C. Xu and J. Prince, “Snakes , shapes, and gradient vector flow," Image Processing, IEEE Transactions on 7, pp. 359-369, Mar 1998.

Segmentation Results Left Maxillary Sinus Right Maxillary Sinus DR GVF DR GVF Front Back

Segmentation Results Errors (mm) as compared to hand segmented ground truth

Segmentation Results Errors (mm) as compared to hand segmented ground truth

Statistical Shape Model (SSM) [8][9] PCA 1. Automatic Segmentation Template Creation Deformable Registration Segmentation Improvement 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [8] T. Cootes, C. Taylor, D. Cooper, and J. Graham, “Active shape models-their training and application ,“ Computer Vision and Image Understanding 61(1), pp. 38-59, 1995. [9] G. Chintalapani, L. M. Ellingsen, O. Sadowsky, J. L. Prince, and R. H. Taylor, “Statistical atlases of bone anatomy: construction, iterative improvement and validation," in Medical Image Computing and Computer- Assisted Intervention, pp. 499-506, 2007.

Statistical Shape Model (SSM) Middle Turbinate Inferior Turbinate Right Maxillary Sinus Left Maxillary Sinus 1 st Mode 2 nd Mode

Correspondence Improvement [12] 1. Automatic Segmentation ∗ Template Creation Deformable Registration Segmentation Improvement 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [12] Seshamani S, Chintalapani G, Taylor RH, Iterative refinement of point correspondences for 3d statistical shape models. MICCAI. 2011;417-425.

Correspondence Improvement [12] 1. Automatic Segmentation ∗ Template Creation Deformable Registration Segmentation Improvement PCA 2. Statistical Shape Modeling Principal Component Analysis Correspondence Improvement [12] Seshamani S, Chintalapani G, Taylor RH, Iterative refinement of point correspondences for 3d statistical shape models. MICCAI. 2011;417-425.

Leave-one-out Analysis Mid iddle Turbinate: Vertex Err Error Mid iddle Turbinate: Res esidual l Sur Surface Err Error 1.6 Iter_0 Iter_0 0.8 Iter_1 Iter_1 1.4 0.7 Iter_2 Iter_2 1.2 Iter_3 Iter_3 mm) 0.6 mm) or (mm or (mm 1 Error 0.5 Error urface Er an Vertex Er 0.8 0.4 al Sur 0.6 Mean Residual 0.3 0.4 0.2 0.2 0.1 0 0 0 10 20 30 40 50 60 0 10 20 30 40 50 60 # # mo modes # mo # modes

Natural Variation • Hypothesis is: Given CTs of 𝑜 individuals, it is likely that the turbinates of each individual are at a different state in the nasal cycle than all others. Therefore, a statistical model of the turbinates built from these 𝑜 CTs should also reflect natural variation.

Natural Variation 𝐽1 = 𝑛 𝑗 𝑈 𝑊 𝐽1 − 𝑐 𝑗 𝑊 Pre-op 𝑜 𝑡 • Experiment CT ∗ = 𝐽1 𝑛 𝑗 𝑊 𝑊 + 𝑐 𝑗 𝐽1 𝑗=1 Compare PATIENT X mode weights 𝐽2 = 𝑛 𝑗 𝑈 𝑊 𝐽2 − 𝑐 𝑗 𝑊 𝑜 𝑡 Post-op ∗ = 𝐽2 𝑛 𝑗 𝑊 𝑊 + 𝑐 𝑗 CT 𝐽2 𝑗=1 • Built separate models for skull and inferior turbinates • We expect inferior turbinates to change, but the skull to not change. This should be reflected in the mode weights when pre-op and post- op models are projected onto our statistical model.

Natural Variation Mode Weig ights: In Inferio ior Turbin inate Mode Weig ights: Bon Bone 4 4 3 3 2 2 ights ts 1 1 ights ts ode weig ode weig 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Mod Mod -1 -1 -2 -2 -3 -3 -4 -4 mode mo mode mo P2a P2b P2a P2b

Natural Variation Difference in Di n Mode Weig eights 5 Bone 4.5 IT IT 4 3.5 3 ifference 2.5 Dif 2 1.5 1 0.5 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 mo mode

Natural Variation

Natural Variation

Population variation vs Natural Variation

Summary • We have built an initial statistical shape model (SSM) of the paranasal sinuses from CT scans of 53 different patients. • SSMs of erectile tissue in the sinuses reflect variations due to the nasal cycle, which are captured in the modes of our PCA models. • A preliminary experiment with a single same-patient pre-op/post-op CT image pair suggests that certain statistical modes are more sensitive than others in characterizing this variation.

Future Work • We are currently working on constructing a larger statistical atlas of the sinuses based on CT scans of 500 patients. • We hope to extend our exploration of the nasal cycle using a larger number of same-patient longitudinal studies. • We are also working to incorporate our results into ongoing research on intraoperative video-CT registration.