1. Motion-adaptive radiotherapy - what to do when the tumor won’t sit still
Martin J Murphy PhD
Department of Radiation Oncology
Virginia Commonwealth University

2. External-beam radiotherapy

3. The nagging problem of external-beam alignment
Except in frame-based radiosurgery, there is no rigid connection between the treatment site and the radiation delivery mechanism
Typically, the treatment beams are aligned to setup landmarks once, at the beginning of irradiation, but
The patient is not a rigid body
The patient is alive and breathing (one would hope)
This leads to
Shape changes in the anatomy and movement of the target during irradiation
Resulting in a time-varying mismatch between the planned dose distribution and the actual anatomy

4. Solutions to the alignment problem

5. The dark ages: irradiate everything

6. The middle ages: Nail the patient to the treatment table, close your eyes, and treat

7. The modern age: Image-guided radiotherapy
Use image-based patient alignment at the beginning of each fraction
Locate the treatment site in the images
Align the beam directly to the treatment site
But, there can still be movement after alignment

8. The paradigm for movement is respiration

9. Respiratory motion Lung tumors can move 3 - 30 mm during respiration
The motion can be spatially and temporally complex and irregular
Breath-holding can be difficult for some patients

10. The most widespread motion adaption method is beam gating
A basic method - turn the beam on and off as the target moves past

11. The post-modern age: Real-time motion-adaptive radiotherapy - follow the moving tumor with the beam
One can move a collimating aperture synchronously with the moving tumor

12. Or, one can move the entire linear accelerator synchronously with the tumor

13. Components of a real-time tumor tracking system
A target localization system to continuously measure the tumor position
A dynamic re-alignment system to shift the beam with respect to the patient, or vice-versa
A real-time control loop to translate tumor position into beam/couch coordinates and synchronously reposition the beam/patient while correcting for delays in the detection/realignment process

14. An example of a real-time tracking system

15. Target localization

16. Direct tumor tracking
Sometimes you get lucky and can clearly see the tumor outline in a 2D radiograph

17. Internal fiducial tracking
Implant radio-opaque markers in or near the tumor before CT
Track the fiducials fluoroscopically

18. The Calypso Electromagnetic localization system

19. Calypso system for electromagnetic tracking
Small electromagnetic coils (transponders) are implanted at the treatment site

20. Calypso electromagnetic tracking
An antenna system with transmitters and receivers is positioned over the patient
The antenna system alternately excites and detects resonant EM radiation from each transponder

21. Surrogate tracking Make a fluoroscopic video or 4D CT of the patient
Correlate the tumor motion with the external marker motion
During treatment, use the external marker motion to predict the tumor position

22. Tumor-chest correlation

23. Tumor-chest correlation coefficient
Good
Bad

24. Dynamic realignment systems

25. Dynamic beam shift - CyberKnife Synchrony
X-ray imaging samples the tumor position
surface markers correlate with the tumor position
Optical tracking of the surface provides continuous tumor position estimates
Real-time control loop synchronizes beam pointing with inferred breathing motion of the tumor

26. Dynamic beam shift - multileaf collimator (video courtesy of Paul Keall, Stanford University)

27. Control loop for synchronization

28. Control loop for synchronization

29. Predictive control loop for system latency
There will be a delay between the time that the tumor position is measured and the time that the beam re-alignment is completed
The delay (latency) can be ms
Therefore the control loop must “lead” the target in order to keep the beam aligned
This requires a real-time predictive algorithm

30. Requirements of a temporal prediction control loop
Accurately predict breathing amplitude up to 500 ms in advance
Accommodate a wide variety of breathing patterns, some of which might be highly irregular
Require little or no customization to individual patients
Continually adjust to breathing that changes with time
Avoid losing track of the breathing pattern during irregular transients
Recover optimal accuracy rapidly following disruptions

31. Methods of prediction
Make a biomechanical model of the breathing process
Make a mathematical model of the breathing cycle using periodic functions
Make a heuristic model of the breathing cycle that mimics each patient’s observed breathing pattern

32. Adaptive filters
Adaptive filters are heuristic algorithms that mimic the incoming signal by taking sequential samples of the signal amplitude and combining them in a weighted sum to estimate the present or future amplitude
The filters make no assumptions about the functional form of the signal or the physical mechanism producing it
The filters can continually adjust their weighting parameters to adapt to changes in the signal shape

33. A basic linear adaptive filter

34. From linear filter to neural network
In a linear filter the output signal estimate is just the weighted sum of the input samples
In a nonlinear filter two or more weighted sums of the input samples are combined in a non-linear function to produce the output estimate
A nonlinear filter can recognize much more complex patterns in the incoming signal than a linear filter

35. A nonlinear neural network

36. Often, breathing is regular and predictable

37. But sometimes it looks like this

38. Summary
Real-time adaptation to respiratory motion is in practice at some clinics
Electromagnetic transponders have been clinically demonstrated to provide accurate real-time position data without x-rays
EM transponders have not yet been combined with real-time beam synchronization - this is a work in progress
Without EM transponders one must either use fluoroscopy to observe the tumor or predict the tumor position from external surrogates
All real-time systems have latencies that must be accommodated by a predictive control loop
Breathing prediction is actually rather difficult and is a very active area of research for real-time respiratory tracking for radiation therapy