Imaging in RT

Question 1

What factors should you consider when optimising imaging doses for RT?

Answer 1

1) For IGRT, produce images with a) dose ALARA and b) adequate image quality for the task
2) Higher doses may, utimately, result in lower doses overall for healthy tissue as improved image quality results in more accurate target localisation, thus sparing healthy tissue from additional (unintended) dose
3) 4D CT - for planning, can be justified if a) produces better quality CT images of moving targets b) allows more accurate measurements of target motion
4) Dose burden dose not change over time
5) Consistent image quality
6) Consistent performance of clinical tasks by trained operators

Question 2

What are the factors that affect image quality and patient dose?

Answer 2

1) Image volume (FOV may be significantly larger than irradiated volume)
a) Detectors energy absorption efficiency

2) Irradiated volume
i. e. image to irradiated volume ratio

Question 3

What does AAPM Report 180 contain?

Answer 3

Tables of organ doses from imaging and guidance.

Question 4

What does AAPM Report TG132 contain?

Answer 4

Information regarding image registration (including evaluation).

Question 5

What does AAPM Report TG179 contain?

Answer 5

Information on imaging dose for various modalities along with clinical applications, QA advice and some information on registration.

Question 6

What factors should you consider when treating a bariatric patient?

Answer 6

1) Can the patient fit into the FOV of the scanner? Typical FOV for CT is 50-60cms; scanner bore 80-85cms.
2) Can the patient external be viewed int he FOV (not just the treatment area)? May need to use smaller arcs if not.
3) Extended FOV - artefacts at extents of field may mean the patient external not clear
4) “CT-ladders” used to determine positioning (metal clips on skin).
5) Does the patient’s weight effect couch movement and is it within the max weight limit of the couch? Limit approx 200-250kgs.
6) Reproducibility of positioning
7) placement of arms - needs for be able to place outside of treatment beam and same position when planning image taken.
8) if large 5-point mask needed, will this fit onto the couch?
9) increased patient size means more patient scatter, signal quality degraded, therefore can increase output from CT scanner, greater current, may have issues with beam on time (tube over-heating); also have max current therefore limit to image quality. Potentially increase slice thickness, reduced resolution in slice direction and could have issues with board placement during treatment. Could also use diagnostic scans (shorter scan) but not in treatment position.
10) may need to overwrite densities (e.g. CT-ladder)
11) Verification protocols - may need to change frequency from weekly to daily as more variation day to day
12) PTV margin increase may be necessary/change tolerance when setting up the patient for treatment

Question 7

What are the requirements for a CT scanner for RT?

Answer 7

1) hard, flat couch top for reproducible positioning
2) compatibility with immobilisation devices
3) lasers for patient positioning
4) high geometric accuracy
5) Accurate map of HU to mass/electron density curves (RMI phantom) for each scan protocol used (as different energies therefore different linear attenuation coeffs); note for pediatric patients (due to different energies used, smaller body) at high mass/electron densities, actual values can deviate from the reference curve.

RT adult head - 120kV (predominant energy around 45kV)
Paediatric head - 80kV (predominant energy 25kV)

Question 8

What is the equation to calculate HU?

Answer 8

1000 x [u - u(water)]/[uwater - uair]

Question 9

From the literature, what are the recommended tolerances for the determination of electron density for:

a) Water
b) lung
c) bone

Answer 9

a) +/- 0.03
b) +/- 0.05
c) +/- 0.08

Question 10

What are typical slice thicknesses for:

a) PET images
b) Planning CT

Answer 10

a) 4mm

b) 1.2mm

Question 11

Why is PET imaging used in RT?

Answer 11

1) to determine lesion characteristics - stageing, metabolic rate, etc
2) assessment to treatment response
3) for trials - don’t need to store image, use generated metrics (lower IT storage requirements)
4) dose optimisation
5) testing drug targeting
6) RT target delineation (functional target volumes, therefore, aiding in reducing inter-observer variability compared with just CT alone)
7) PET can distinguish between tumour (metabolically active) and non-tumour (relatively metabolically inactive) masses, e.g. atelectasis in the lung.

Note: may have to take concomitant dose from PET imaging into consideration for RT plan.

Question 12

What is the attenuation correction?

Answer 12

This corrects for the attenuation due to the patient from a coincident event (C) where the two annihilation photons are simultaneously detected.

C = C0 x exp[-uD]

C0 =  attenuated signal detected
D = total thickness through body/tissue
u = linear attenuation coefficient

Question 13

What is SUV?

Answer 13

Standard Uptake Value attempts to normalise uptake in tissues to the patient’s body weight.

SUV = activity concentration (Bq/ml) / [injected dose (Bq) / Body weight (g)]

Note an additional factor of: Glucose level in plasma (mmol/litre)/5.0 (mmol/litre), can be added to compensate for the uptake of glucose rather than FDG.

Question 14

PET, SUV: what options could be used for body mass/weight?

Answer 14

1) Body mass/weight - most common as easy to calcuation but can be influenced by body shape
2) Lean body mass - more complex to calculate and there are different measurement/estimation techniques but lean body mass is more consistent across a range of body ‘habitus’ (body shape) for FDG; FDG very low uptake in fat as not as metabolic as muscle.
3) Body surface area - tallies with some dosing regimes

Question 15

What are the three types of SUV values used for PET and what are their characteristics?

Answer 15

1) SUVmax: max value within an ROI, most commonly used as easy to determine but is dependent on the ROI drawn (inter-observer variation) and noise in the image can skew interpretation as making judgements based on a single pixel
2) SUVmean: mean value within an ROI; depends on ROI, therefore inter-observer variability but is less susceptible to noise than SUVmax
3) SUVpeak: uses small, fixed volume for an ROI and calculates the highest value; therefore, regional values do not affect it so it represents local higher values; this is a compromise between the other two types; ROI placement still varies with clinician but not volume; reduced noise influence; ROI volume and shape needs to be standardised if being used for comparisons.

Question 16

Calibration of a PET scanner.

Answer 16

For absolute dosimetry, the scanner is cross-calibrated to a traceable dose calibrator (well chamber) to ensure SUV measurements remain within a specific tolerance.

This is conducted quarterly at Leeds; uniform phantom (uniformity test) containing water + approx 20MBq of FDG (evenly mixed and measured by the traceable dose calibrator) is scanned for 5 minutes.

An ROI over the whole slice is placed on a selection of slices and the generated SUV value is 1.00 +/-0.1; a Ge phantom is also scanned, an ROI is placed over the whole slice for several slices and the SUV value recorded and checked to be 1.00+/-0.1

I believe these tolerances are not only in the literature but are from the CORE programme which ensures scanners from different centres have comparable image qualities to facilitate comparisons between centres.

Question 17

Concomitant dose: when comparing national dose reference levels (e.g. IPEM National Dose Audit) what factors should you consider?

Answer 17

1) are the protocols at your centre different?
2) Are the differences appropriate?
3) do you require improved image quality due to performing advanced/difficult techniques?
4) Do you monitor per patient or retrospective audit?

Note: workflow/throughput is much higher in RT than in DR.

Question 18

For IGRT, in terms of optimisation, explain two factors that should be considered.

Answer 18

1) Doses from imaging equipment need to be ALARA and the dose burden does not change over time
2) Image quality is sufficient for the required task and does not change over time

Note: higher IGRT doses may, overall, result in a net dose saving to healthy tissue as greater quality images may lead to more accurate localisation and greater sparing of healthy tissue.

Additionally, for kV imaging, the imaged volume and irradiated volume differ with the irradiated volume being greater than the imaged volume. This difference depends on the individual system.

Question 19

What main detector factor affects patient dose and image quality?

Answer 19

Energy absorption efficiency

Question 20

For IGRT, if you wanted to lower the overall risk of secondary induced cancers (e.g. for a radiosensitive OAR in the irradiated FOV) what potentially could you do?

Answer 20

1) alter the field of view/scan length to exclude or partially exclude the OAR (care taken not to then irradiate other radiosensitive OAR)
2) Lower mAs whilst still achieving sufficient image quality (test a range of mAs on phantom scans); examine noise (standard deviation of HU) and work out the minimum mAs required to give sufficient image quality (subjective assessment by radiographers and clinicians).
3) lower the frequency of IGRT - this would require MDT assessment and a risk assessment evaluating risk of localisation error

Note, risk of death from surgery is approx 1/100, secondary induced cancer risk for prostate patients given in example was 1/160 and 1/960 for the two systems.

Also, AAPM180 recommends if IGRT/concomitant dose is >5% therapeutic dose (from CHART study) it must be taken account in the treatment plan.

Question 21

What are the 4 ‘ingredients’ in a registration algorithm?

Answer 21

1) Metric to measure image similarity
2) Transform to act on the moving image to map it to the fixed image
3) Optimiser to find the transform that optimises (min/maximises) the metric
4) Interpolator to resample the moving image during the transformation

Question 22

When registering an image, what are 3 possible interpolation algorithms that could be used?

Answer 22

1) Nearest Neighbour
2) linear interpolation
3) Nth order B-spline

Question 23

What are the characteristics of Mean Square Difference as a metric for image registration?

Answer 23

1) only used on unimodal registrations (e.g. CT to CT)
2) requires the 2 images to have the same image intensity distribution
3) very sensitive to small number of voxels with large intensity differences

Note MSD is not used clinically.

Question 24

What are the characteristics of Mutual Information (MI) as a metric for image registration?

Answer 24

1) most common registration metric
2) intensities don’t have to be the same
3) multi-modal (e.g. MRI to CT)
4) is uses a measurement of entropy

Question 25

What is a kernel?

Answer 25

A matrix of pixel weighting factors applied to each pixel in an image using a convolution.

Question 26

What 3 types of registration are there?

Answer 26

1) rigid - translation (3 degrees of freedom) or translation + rotation (6 degrees of freedom)
2) Affine - rigid + scaling and shearing (12 degress of freedom)
3) Deformable - no geometric relationship in moving image preserved (n degrees of freedom)

Question 27

Outline the process of image segmentation using atlas-based methods.

Answer 27

1) have an atlas of images of the same modality with “expert” contours
2) acquire new images of the same modality
3) deformably register each atlas image to the new image

4) transfer atlas contours to new image using the same deformation transform
5) combined the contours to produce one final contour set

Question 28

What are the issues associated with atlas-based segmentation?

Answer 28

1) need a large number of patients in the database to make it robust
2) it cannot deal with unusual anatomy
3) requires images to be of similar current image e.g. contrast-enhanced
4) may require significant editing, therefore negating its benefit
5) Depends on accuracy of contour in the atlas

Question 29

Outline the process for model-based image segmentation.

Answer 29

1) create a 3D average model of the organ using an atlas of image sets and “expert” contours
2) extract the typical shape variation of the organ from the average model using the atlas
3) model is initialized using deformable registration to the ‘average’ atlas image set
4) an adaptation algorithm then seeks to apply the appropriate shape variation in the model to best fit the image data using image processing techniques

Question 30

What are the issues associated with model-based segmentation?

Answer 30

1) depends on the accuracy of the model
2) in theory independent of image modality, although more dependent on image quality
3) model may be different from local contouring practice, e.g. different approach to contouring
4) may still require significant editing, therefore not time conserving

Question 31

Outline the process of deep learning segmentation.

Answer 31

1) have an atlas of images with “expert” contours (greater number = more robustness)
2) setup algorithm network architecture 3D convolution Neural Network (U-NET as it has a U shaped structure to capture data at different resolutions)
3) Train model using atlas data
4) input new image, codel creates contours (very rapid, <1min)

Question 32

What are the issues with deep learning segmentation?

Answer 32

1) depends on the quality of manual contouring
2) requires large amount of training data to produce reliable results (100+ patients)
3) Training times very long
4) ‘blackbox’ hard to predict how it will behave in unusual situations

Question 33

How does thresholding work and give an example of its use clinically?

Answer 33

Windows to effectively 0 width, i.e. displaying binary values. Useful when there are large intensity changes in the image such as the lung to surrounding tissue; here thresholding is used to quickly contour the lung in treatment planning.

Question 34

In terms of verification of patient treatment, what 2 types of random error are there? What are their possible causes?

Answer 34

1) individual - standard deviation of the measured errors over the course of the treatment for an individual patient
2) population - mean for the individual random errors for each patient

Possible causes:

a) patient setup error
b) target position and shape
c) intrafraction errors

Question 35

List some verification techniques (where RT geometric accuracy is assessed).

Answer 35

1) Pre-treatment verification - compares ref images with planned treatment BEFORE treatment course started
2) Off-line treatment verification - compares reg images with images taken in the treatment room and analyses set-up accuracy AT SOME TIME AFTER treatment
3) Online treatment verification - on every fraction, compare reference images with treatment images than in treatment room IMMEDIATELY PRIOR TO treatment
4) Real-time treatment merfication - compares ref images with images taken in treatment room AS TREATMENT IS DELIVERED (e.g. gated delivery)

Question 36

In terms of geometric uncertainties, which publications provide guidance?

Answer 36

Geometric uncertainties in RT (BIR, 2003)
Ontarget: ensuring geometric accuracy in RT (RCR, 2008)

Note, On target 2 is published June 2021 (think this is to replace both of these)