Algorithmic Challenges in Radiation Therapy Guillaume Blin January, - - PowerPoint PPT Presentation

▶

algorithmic challenges in radiation therapy

Algorithmic Challenges in Radiation Therapy Guillaume Blin January, - - PowerPoint PPT Presentation

Feb 05, 2023 325 likes •992 views

Algorithmic Challenges in Radiation Therapy Guillaume Blin January, 2019 Complexity, Algorithms, Automata and Logic Meet 2019 Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 1 / 28 Radiation Therapy Radiation

slide-1

SLIDE 1

Algorithmic Challenges in Radiation Therapy

Guillaume Blin January, 2019 Complexity, Algorithms, Automata and Logic Meet 2019

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 1 / 28

slide-2

SLIDE 2

Radiation Therapy

Radiation Therapy Cancer treatment relying on radiations aiming at killing cancerous cells.

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 2 / 28

slide-3

SLIDE 3

Therapy modalities

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 3 / 28

slide-4

SLIDE 4

External beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 4 / 28

slide-5

SLIDE 5

External beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 4 / 28

slide-6

SLIDE 6

External beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 4 / 28

slide-7

SLIDE 7

External beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 4 / 28

slide-8

SLIDE 8

External beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 4 / 28

slide-9

SLIDE 9

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-10

SLIDE 10

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-11

SLIDE 11

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-12

SLIDE 12

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-13

SLIDE 13

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-14

SLIDE 14

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-15

SLIDE 15

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-16

SLIDE 16

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-17

SLIDE 17

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-18

SLIDE 18

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-19

SLIDE 19

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-20

SLIDE 20

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-21

SLIDE 21

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-22

SLIDE 22

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-23

SLIDE 23

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-24

SLIDE 24

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-25

SLIDE 25

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-26

SLIDE 26

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-27

SLIDE 27

Internal beam therapy

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 5 / 28

slide-28

SLIDE 28

Different particles

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 6 / 28

slide-29

SLIDE 29

Pros and Cons

Figure 1: Taheri-Kadkhoda et al. Radiation Oncology 2008 Figure 2: UCLA Brachytherapy Program

Protons and brachy therapies spare more healthy tissues

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 7 / 28

slide-30

SLIDE 30

Pros and Cons

Figure 1: Protons center are expensive - 95 Millions euros, size of a building

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 7 / 28

slide-31

SLIDE 31

Pros and Cons

Figure 1: Bragg peak Figure 2: Motion sensitivity

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 7 / 28

slide-32

SLIDE 32

Pros and Cons

Figure 1: Brachy is invasive and needs catheter or needles to reach the tumor site

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 7 / 28

slide-33

SLIDE 33

Main common problem

Take into account specificities between patients or along the treatment for a single one due to variance arising in

Patient setup Patient breathing / coughing Patient heart-beat Patient discomfort Patient weight fluctuation Patient implants ...

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 8 / 28

slide-34

SLIDE 34

What about the algorithmic in all this ?

Binary matrices Stringology Pathways in graph Big data Deep learning

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 9 / 28

slide-35

SLIDE 35

Multileaf collimators

Optimize total and/or setup time

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 10 / 28

slide-36

SLIDE 36

Multileaf collimators

Minimizing the total beam-on time is solvable in linear time Minimizing the total setup time is Strongly NP-hard even for matrices with a single row We investigated algorithmic aspects of two technological variants in Sofsem 2014

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 11 / 28

slide-37

SLIDE 37

Multileaf collimators variants

Figure 2: Rotating Collimator

The Rotating MLC Decomposition problem is NP-Hard when minimizing either the total setup time or the total beam-on time Approximable with an additional overcost relative to size

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 12 / 28

slide-38

SLIDE 38

Multileaf collimators variants

Figure 2: Multi-Layer Multileaf Collimator

The Dual-MLC Decomposition problem is NP-Hard when minimizing the total setup time.

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 12 / 28

slide-39

SLIDE 39

Modulated brachytherapy

Figure 3: Modulated brachytherapy

Conformation to the shape of the tumor site In practice, computation are done relatively to dose absorption in water

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 13 / 28

slide-40

SLIDE 40

Tunable shield

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 14 / 28

slide-41

SLIDE 41

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-42

SLIDE 42

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-43

SLIDE 43

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-44

SLIDE 44

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-45

SLIDE 45

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-46

SLIDE 46

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-47

SLIDE 47

Tunable shield

Circular integer word decomposition into circular binary words under constraints

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 15 / 28

slide-48

SLIDE 48

Tunable shield

The shield configuration can be considered as fixed or dynamic Provided with or without rotation capabilities Allowing or not irradiation overdoses We investigated algorithmic aspects of those variants in IWOCA 2016

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 16 / 28

slide-49

SLIDE 49

Known results

Provided with one single fixed configuration Allowing overdose, it can be solved in O(N log N). Forbidding overdose, it can be solved in O(N)

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 17 / 28

slide-50

SLIDE 50

Known results

Considering multiple shield configurations allowed Achieving the optimal difference between the prescribed dose and the actual total delivered dose using a minimal number of shield configurations Given an upper bound on the number of shield configurations, achieving the minimum reachable difference Both are NP-hard even when each shield sector is associated to a even number of consecutive patient volumes But can be approximated in polynomial time within a factor of log of the max prescribed dose of the optimum

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 18 / 28

slide-51

SLIDE 51

Customized cylindrical shields for brachytherapy

Manufacturing a given single best shield for a given patient (3D Metal printing) Assume that the physical precision of

ur process is limited (lower bounds on

the size of a closed or open sector of a produced shield) We investigated algorithmic aspects of the corresponding problem in CPM 2018

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 19 / 28

slide-52

SLIDE 52

Customized cylindrical shields for brachytherapy

Given a circular integer word w, the cylindrical shield to be designed can be seen as a constrained circular binary word of the same length where, when we replace each 1 by the selected irradiation time t, the Manhattan distance to w is minimal. Constraints on the circular binary word are according to the minimal length for an opening, and for a closed sector between two openings

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 20 / 28

slide-53

SLIDE 53

Customized cylindrical shields for brachytherapy

A pseudo-polynomial time algorithm of complexity O(|w| ∗ tmax ∗ l3) exists with tmax the maximal time of an irradiation and l the maximum sector size w = 013331102230313210 with l0 = 3, l1 = 5

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 21 / 28

slide-54

SLIDE 54

Pencil Beam Discrete Scanning

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 22 / 28

slide-55

SLIDE 55

Pencil Beam Discrete Scanning

The beam is turned off between the spot positions

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 22 / 28

slide-56

SLIDE 56

Pencil Beam Discrete Scanning

The beam is turned off between the spot positions

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 22 / 28

slide-57

SLIDE 57

Treatment Planning

Plan: ... Step k: Energy, x coordinate , y coordinate, duration ...

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 23 / 28

slide-58

SLIDE 58

Treatment Planning

Plan: ... Step k: Energy, x coordinate , y coordinate, duration ...

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 23 / 28

slide-59

SLIDE 59

Treatment Planning

Plan: ... Step k: Energy, x coordinate , y coordinate, duration ...

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 23 / 28

slide-60

SLIDE 60

Treatment Planning

Plan: ... Step k: Energy, x coordinate , y coordinate, duration ...

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 23 / 28

slide-61

SLIDE 61

Treatment Planning

Plan: ... Step k: Energy, x coordinate , y coordinate, duration ...

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 23 / 28

slide-62

SLIDE 62

Optimization of paths

Not so much investigated from the algorithmic point of view Necessity to take into account motion sensitivity We proposed an ”An

pen-source motion simulator

for proton therapy algorithmic aspects” (MSPT) in the PhD thesis of Paul Morel 2015 which reflects the consequences of motion on a treatment

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 24 / 28

slide-63

SLIDE 63

What about big data and deep learning ?

How to provide personal treatment plans in real time like fashion ? Take advantage from past treatment plans

Gathering treatment plans Storing them in an efficient way Query them in real time

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 25 / 28

slide-64

SLIDE 64

What about big data and deep learning ?

How to provide personal treatment plans in real time like fashion ? Take advantage from past treatment plans

Avoid redundant computation Being able to start from a realistic draft of the treatment plan rather than from scratch Compute alternative plans to react in real time

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 26 / 28

slide-65

SLIDE 65

Learning plans rather than computed them

Underline physics is complicated Rather than trying to compute it faster, could one learn it somehow ?

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 27 / 28

slide-66

SLIDE 66

Conclusion

Conclusion

Radiation therapies provides lots of interesting problems One can easily find its own algorithmic playground

Guillaume Blin Algorithmic Challenges in Radiation Therapy January, 2019 28 / 28