Endoscopic navigation in the absence of CT imaging Ayushi shi Sinha ha 1, , , Xingtong Liu 1 , Austin Reiter 1 , Masaru Ishii 2 , Greg Hager 1 , Russ Taylor 1 1 The Johns Hopkins University, Baltimore, USA 2 Johns Hopkins Medical Institutes, Baltimore, USA Code: https://gi tps://github hub.c .com/ m/Ayus yushi hiSinha/ci Sinha/cisstICP tICP sinha nha@j @jhu hu.edu edu
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011 Navi viga gati tion on withou hout addi diti tional onal to tools ols
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011 Navi viga gati tion on withou hout addi diti tional onal to tools ols Estimat mate e anat natomy omy without hout CT T scan an
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011 Navi viga gati tion on withou hout addi diti tional onal to tools ols Estimat mate e anat natomy omy without hout CT T scan an Assign ign confi fiden dence ce to to registr gistration ion
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011 Navi viga gati tion on withou hout addi diti tional onal to tools ols Estimat mate e anat natomy omy without hout CT T scan an Assign ign confi fiden dence ce to to registr gistration ion
Navigation without additional tools Learning-based method Liu, X. et al., "Self-supervised Learning for Dense Depth Estimation in Monocular Endoscopy", CARE Workship 2018
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011 Navi viga gati tion on withou hout addi diti tional onal to tools ols Estimat mate e anat natomy omy without hout CT T scan an Assign ign confi fiden dence ce to to registr gistration ion
Estimate anatomy without CT scan • Build statistical shape models • Principal component analysis • Capture anatomical variation • Deformable registration • Optimize PCA model parameters • Produce registration score
Estimate anatomy without CT scan • Build statistical shape models • Principal component analysis • Capture anatomical variation • Deformable registration • Optimize PCA model parameters • Produce registration score
Statistical shape models • Given shapes, , with correspondences, we can compute: • Variance: • Mean:
Statistical shape models • Variance along the principal mode for middle turbinates
Statistical shape models • Given a new shape, , we can compute: • Mode weights: • Estimated shape:
Estimate anatomy without CT scan • Build statistical shape models • Principal component analysis • Capture anatomical variation • Deformable registration • Optimize PCA model parameters • Produce registration score
Deformable most likely point (D-IMLP) Find R, t and a such that x is best aligned with a deformed y … Find s such that y deforms to fit x y 𝑗 ∈ Ψ X = {x 𝑗 } 𝑔(y i , s)
Generalized deformable most likely oriented point (GD-IMLOP) Find R, t and a such that x is best aligned with a deformed y … and such that the normal of y aligns with that of x Find s such that y deforms to fit x y 𝑗 ∈ Ψ X = {x 𝑗 } 𝑔(y i , s) , ,
What is ? (2) 𝐰 𝑗 (3) 𝜈 𝑗 (1) 𝐰 𝑗 y 𝑗 (1) 𝜈 𝑗 (2) 𝜈 𝑗 (3) 𝐰 𝑗 18
Nasal endoscopy in the clinic G. Scadding et al., Diagnostic tools in Rhinology EAACI position paper , Clinical and Translational Allergy, 1(2), 2011 Navi viga gati tion on withou hout addi diti tional onal to tools ols Estimat mate e anat natomy omy without hout CT T scan an Assign ign confi fiden dence ce to to registr gistration ion
Did it work? ≈ 𝜓 2 distribution = ≈ 𝜓 2 = distribution sity ty ity densi 𝑞 = Pr[E 𝑞 < 𝜓 2 ] abilit Probabil 𝜓 2 Chi-squar square e value lue
Did it work? ≤ = ≤ = sity ty ity densi 𝑞 = Pr[E 𝑞 < 𝜓 2 ] abilit Probabil 𝜓 2 Chi-squar square e value lue
Experiments
Leave-one-out • # sample point: 3000 • Translational offset: [0, 10] mm • Rotational offset: [0, 10] degrees • Noise: • 0.5 × 0.5 × 0.75mm 3 • 10° (𝑓 = 0.5) • Noise assumed: • 1 × 1 × 2mm 3 • 30° (𝑓 = 0.5) • 𝑜 𝐧 ∈ {0, 10, 20, 30, 40, 50} Right nasal airway
Leave-one-out
Leave-one-out
Leave-one-out
Leave-one-out 𝑞 = 0.95 (ve very confident) t) TR TRE = 0.34 (± 0.03) mm mm
Leave-one-out 𝑞 = 0.9975 (confid ident) t) TR TRE = 0.62 (± 0.03) mm mm
Leave-one-out 𝑞 = 0.9999 (some mewhat hat confident) t) TR TRE = 0.78 (± 0.04) mm mm
Leave-one-out 𝑞 = 0.999999 (low confidence) ce) TR TRE = 0.80 (± 0.05) mm mm
Leave-one-out remain inin ing (n (no o confidence) e) TRE > 1 mm
In vivo • 5 clinical sequences • 3000 sample points • Noise assumed: • 1 × 1 × 2mm 3 • 30° (𝑓 = 0.5) • 𝑜 𝐧 ∈ {0, 10, 20, 30, 40, 50} Right nasal airway Dense reconstruction from video
In vivo Residual Max Min # error error error registrations (mm) (mm) (mm) All registrations 30/30 1.09 (±1.03) 4.74 0.50 Registrations that pass E p test 27/30 0.76 (±0.14) 0.99 0.50 Registrations that pass E p and E o tests 12/30 0.78 (±0.07) 0.94 0.72
Conclusions and future work • Navigation without additional tools • Estimate anatomy without CT scan • Assign confidence to registration • Learn statistics from 1000s of CTs • Use additional features • Evaluate further on in vivo data
Thank you! Code de: https://github.com/AyushiSinha/cisstICP Poster er: W-2 Acknow nowle ledgemen dgements ts: This work was funded by NIH R01-EB015530, NSF Graduate Research Fellowship Program, an Intuitive Surgical, Inc. fellowship, and JHU internal funds. : : sinha@jhu.edu : @ItsAyushiSinha
Recommend
More recommend
