Personalization of Pediatric Imaging in Terms of Needed Indication-Based Quality Per Dose Acknowledgments Duke University Medical Center Ehsan Samei, PhD Donald Frush, MD Xiang Li PhD DABR Cleveland Clinic Formerly, Duke University Medical Center 2 A Specific Diagnostic Task: Lung Nodule Detection Children with (extra-thoracic) solid tumors are at risk for pulmonary metastatic disease, manifested as lung nodules. Small size (3-5 mm) low contrast (200-500 HU) A Specific Diagnostic Task: Lung Nodule Detection Presence of a single nodule affects treatment plan and prognosis Chest CT to screen for lung nodules at Many CT scans! > initial evaluation > throughout treatment > during routine follow-up Curable cancers long life expectancy Small size (3-5 mm) low contrast (200-500 HU) Easily missed under high-noise conditions require low dose require high dose 3 4 Mean Age (year) Chest CT s GE LightSpeed VCT scanner kvp Scan FOV Slice Thickness Collimation Pitch Noise Max Index ma pink 0.2 00 Ped Body 2.5 20 0.969 9.5 70 red 0.7 00 Ped Body 20 0.969 0 80 purple.4 00 Small Body 20 0.969 0.5 90 yellow 2.5 20 Small Body 5 20.375 70 white 4. 20 Small Body 5 20.375 2 75 blue 6.2 20 Medium Body 5 20.375 3 80 orange 8.2 20 Medium Body 5 20.375 4 90 green.3 20 Medium Body 5 40.375 5 95 black 4.7 20 Large Body 5 40.375 6 0 5 Mean Age (year) Chest CT s GE LightSpeed VCT scanner kvp Scan FOV Slice Thickness Collimation Pitch Noise Max Index ma pink 0.2 00 Ped Body 2.5 20 0.969 9.5 70 red 0.7 00 Ped Body 20 0.969 0 80 purple.4 00 Small Body 20 0.969 0.5 90 yellow 2.5 20 Small Body 5 20.375 70 white 4. 20 Small Body 5 20.375 2 75 blue 6.2 20 Medium Body 5 20.375 3 80 orange 8.2 20 Medium Body 5 20.375 4 90 green.3 20 Medium Body 5 40.375 5 95 black 4.7 20 Large Body 5 40.375 6 0 intended for general diagnostic purposes, but not tailored to a specific task 6
Task s consistent diagnostic accuracy? Step 2 How to optimize for a specific task? Step 3 7 8 How can we obtain this relationship quicker? Monte Carlo study CTDI vol or SSDE Study Study 2 Study 3 E / DLP exp( d ) k chest k 9 Li et al., Med Phys, 38(), 397-407 (20) Li et al., Radiology, 259, 862-874 (20) 0 lung nodule detection: () nodule contrast (2) nodule size (3) noise in the lung How can we obtain this relationship quicker? Data from CT manufacturers? AAPM TG 233: (diameter,scan parameters) f (noise index) Li X et al., Med Phys. 38(5):2609-8 (20). Solomon JB et al., AJR Am J Roentgenol. 200(3):592 600 (203). Image Quality = f (Diameter, Scan Parameters) Wilson et al., Medical physics 40 (3): 03908, 203 2 2 2
diagnostic accuracy: area under ROC curve (AUC) image quality: CDNR display Contrast Displayed Diameter Noise ROC observer study real images + simulated nodules + simulated noise noise = 0 HU noise = 4 HU noise = 20 HU b y a cexp( d log x) Li et al. Br J Radiol. 82(977):40- (2009) Li et al. Med Phys 38(5), 2609-268 (20) Li et al. Med Phys 38(5), 2609-268 (20) How can we obtain this relationship quicker? model observers Study Study 2 Study 3 Yu et al., Med. Phys. 207. doi:0.002/mp.2380 Barrett et al. Physics in medicine and biology 60.2 (205): R. Solomon et al., Proc. SPIE. Vol. 946. 205 6 AUC=0.89 AUC=0.67 7 8 3 3
average AUC=0.83 Step 2 9 20 no iodinated contrast used low incidence of calcification Nodules low-density soft tissue For the purpose of detecting lung nodules, choice of kvp is unimportant. 2 22 Soft-tissue contrast is superior at a lower kvp How about using 80 kvp for all pediatric patients? require higher ma limitations of low kvp exceed system limit increase rotation time decrease pitch more beam-hardening artifact 80 kvp tube voltage (kvp) category weight (kg) Singh et al. Yu et al. our choice pink 5.5-7.4 80 80 80 red 7.5-9.4 80 80 80 purple 9.5-.4 00 00 00 yellow.5-4.4 00 00 00 white 4.5-8.4 00 00 00 blue 8.5-23.4 00 00/20 00 orange 23.5-29.4 00/20 20 00 green 29.5-36.4 20 20 20 black 36.5-55 20 20 20 prolong scan time more motion artifact Cody et al., AJR, 2004 23 Singh et al. Radiology 2009;252():200-208 Yu et al. RadioGraphics 20; 3:835 848 24 4 4
Optimize Slice Thickness Optimize Slice Thickness Mean Slice Age Thickness (year) Small Size (3-5 mm) pink 0.2 2.5 red 0.7 purple.4 yellow 2.5 5 white 4. 5 blue 6.2 5 orange 8.2 5 green.3 5 black 4.7 5 t 25 5-mm 4-mm partial volume averaging 26 Optimize Slice Thickness current Mean Slice Age Thickness (year) pink 0.2 2.5 red 0.7 purple.4 yellow 2.5 5 white 4. 5 blue 6.2 5 orange 8.2 5 green.3 5 black 4.7 5 optimized Slice Thickness 2.5 Mean Age (year) kvp Scan FOV (bowtie filter) Slice Collimation Thickness Pitch pink 0.2 80 Ped Body (s) 2.5 20.375 red 0.7 80 Ped Body (s) 20.375 purple.4 00 Small Body (s) 20.375 yellow 2.5 00 Medium Body (m) 20.375 white 4. 00 Medium Body (m) 20.375 blue 6.2 00 Medium Body (m) 20.375 orange 8.2 00 Medium Body (m) 20.375 green.3 20 Medium Body (m) 40.375 black 4.7 20 Medium Body (m) 40.375 Maximize dose efficiency Achieve desired spatial resolution Respect system limit and scanning speed requirement 27 28 Step 2 Step 3 29 Samei et al., J. Med. Imag. (207) 30 5 5
Breathing Artifact 0.9 msv 4-month old AUC=0.83 Not factored into image quality metric: CDNR display Contrast Displayed Diameter Noise 2 msv Slightly blurred blood vessels Higher lung density (less air in lung due to shallow breathing) 3 32 Conclusions A framework for formulating size-specific pediatric protocols for a given diagnostic task Diagnostic Accuracy = f (Radiation Dose) To achieve the same diagnostic accuracy, more dose is needed to scan a larger/older patient Optimize individual scan parameters Achieve consistent diagnostic accuracy across pediatric sizes 33 Thank you! 35 6 6