Selection of a subset of atlases from a set of candidate atlases

1. A method of selecting from a set of N candidate atlases at least one subset of M candidate atlases for use by at least one medical image auto-contouring system; the method comprising:

for each candidate atlas n within the set of N candidate atlases, computing at least one auto-contouring performance measure in relation to each training atlas t within at least a subset of a set of T training atlases;

for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed, deriving a score based at least partly on the respective auto-contouring performance measure;

for each candidate atlas n within the set of N candidate atlases, aggregating the derived scores for each pair combination for which an auto-contouring performance measure has been computed for that candidate atlas n; and

selecting at least one subset of M candidate atlases from the set of N candidate atlases based at least partly on the most favourable aggregated scores therefor.

2. The method of claim 1 , wherein the score derived for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed comprises at least one of:

a score directly derived from the auto-contouring performance measure;

a relative weighting score derived with respect to all pair combinations of candidate atlases n and a common training atlas t for which an auto-contouring performance measures have been computed;

a relative weighting score derived with respect to a number of training atlases for which at least one auto-contouring performance measure has been computed for the candidate atlas n of the respective pair combination;

an absolute weighting score for the auto-contouring performance measure; and an editing time prediction value.

3. The method of claim 1 , wherein the method comprises computing auto-contouring performance measures for each of a plurality of structures of each candidate atlas n within the set of N candidate atlases in relation to each training atlas t within the respective subset of training atlases.

4. The method of claim 3 , wherein the method comprises, for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed, deriving a score for each of the plurality of structures within the candidate atlas n based at least partly on the computed auto-contouring performance measures, and where the score is based at least partly on a structure weighting value.

5. The method claim 4 , wherein the method comprises, for each candidate atlas n within the set of N candidate atlases, aggregating the derived scores for the plurality of structures for each pair combination of candidate atlas and training atlas comprising that candidate atlas n and a training atlas t for which auto-contouring performance measures have been computed, and selecting the at least one subset of M candidate atlases from the set of N candidate atlases based at least partly on the aggregated scores for all structures for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed.

6. The method of claim 5 , wherein the method further comprises determining whether at least one further subset of M candidate atlases is to be selected from the set of N candidate atlases, and if it is determined that a further subset of M candidate atlases is to be selected from the set of N candidate atlases:

removing the candidate atlases within the previously selected subset of M candidate atlases from the set of N candidate atlases to create a new set of (N-M) candidate atlases;

and/or removing from the set of T training atlases those training atlases T′ for which the candidate atlases within the previously selected subset of M candidate atlases achieved a favourable auto-contouring performance to create a new set of (T-T′) training atlases;

for each candidate atlas n within the new set of (N-M) candidate atlases, computing at least one auto-contouring performance measure in relation to each training atlas t within at least a subset of the new set of (T-T′) training atlases;

for each pair combination of candidate atlas n from the new set of (N-M) candidate

atlases and training atlas t for which an auto-contouring performance measure has been computed, deriving a score based at least partly on the respective auto-contouring performance measures;

for each candidate atlas n within the new set of (N-M) candidate atlases, aggregating the derived scores for each pair combination for which an auto-contouring performance measure has been computed for that candidate atlas n; and

selecting at least one further subset of M candidate atlases from the new set of (N-M) candidate atlases based at least partly on the aggregated scores therefor.

7. The method of claim 5 , wherein auto-contouring performance measures are computed for either:

different candidate atlases within the set of N candidate atlases in relation to different subsets of training atlases from the set of T training atlases;

or for all candidate atlases within the set of N candidate atlases in relation to the same subset of training atlases from the set of T training atlases.

8. The method of claim 5 , wherein the set of T training atlases either comprises of the set of N candidate atlases or the set of N training atlases less the candidate atlas n.

9. An atlas management system comprising at least one processor module arranged to select from a set of N candidate atlases at least one subset of M candidate atlases for use by at least one medical image auto-contouring system; the at least one processor module is arranged to:

for each candidate atlas n within the set of N candidate atlases, compute at least one auto-contouring performance measure in relation to each training atlas t within at least a subset of a set of T training atlases;

for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed, derive a score based at least partly on auto-contouring performance measures comprising at least one of:

a score directly derived from the auto-contouring performance measure;

a relative weighting score derived with respect to all pair combinations of candidate atlases n and a common training atlas t for which an auto-contouring performance measures have been computed;

a relative weighting score derived with respect to a number of training atlases for

which at least one auto-contouring performance measure has been computed for the candidate atlas n of the respective pair combination;

an absolute weighting score for the auto-contouring performance measure; and

an editing time prediction value

for each candidate atlas n within the set of N candidate atlases, aggregate the derived

scores for each pair combination for which an auto-contouring performance measure has been computed for that candidate atlas n; and

select at least one subset of M candidate atlases from the set of N candidate atlases based at least partly on the aggregated scores therefor.

10. The atlas management system of claim 9 , wherein the at least one processor module is arranged to compute auto-contouring performance measures for each of a plurality of structures of each candidate atlas n within the set of N candidate atlases in relation to each training atlas t within the respective subset of training atlases.

11. The atlas management system of claim 10 , wherein the at least one processor module is arranged to, for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed, derive a score for each of the plurality of structures within the candidate atlas n based at least partly on the computed auto-contouring performance measures, and where the score is based at least partly on a structure weighting value.

12. The atlas management system of claim 10 , wherein the at least one processor module is arranged to, for each candidate atlas n within the set of N candidate atlases, aggregate the derived scores for the plurality of structures for each pair combination of candidate atlas and training atlas comprising that candidate atlas n and a training atlas t for which auto-contouring performance measures have been computed, and select the at least one subset of M candidate atlases from the set of N candidate atlases based at least partly on the aggregated scores for all structures for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed.

13. The atlas management system of claim 10 , wherein the at least one processor module is further arranged to determine whether at least one further subset of M candidate atlases is to be selected from the set of N candidate atlases, and if it is determined that a further subset of M candidate atlases is to be selected from the set of N candidate atlases:

remove the candidate atlases within the previously selected subset of M candidate atlases from the set of N candidate atlases to create a new set of (N-M) candidate atlases;

and/or remove from the set of T training atlases those training atlases T′ for which the candidate atlases within the previously selected subset of M candidate atlases achieved a favourable auto-contouring performance to create a new set of (T-T′) training atlases;

for each candidate atlas n within the new set of (N-M) candidate atlases, compute at least one auto-contouring performance measure in relation to each training atlas t within at least a subset of the new set of (T-T′) training atlases;

for each pair combination of candidate atlas n from the new set of (N-M) candidate atlases and training atlas t for which an auto-contouring performance measure has been computed, derive a score based at least partly on the respective auto-contouring performance measure;

for each candidate atlas n within the new set of (N-M) candidate atlases, aggregate the derived scores for each pair combination for which an auto-contouring performance measure has been computed for that candidate atlas n; and

select at least one further subset of M candidate atlases from the new set of (N-M) candidate atlases based at least partly on the aggregated scores therefor.

14. A non-transitory computer program product having executable program code stored therein for selecting from a set of N candidate atlases at least one subset of M candidate atlases for use by at least one medical image auto-contouring system, the program code operable for:

for each candidate atlas n within the set of N candidate atlases, computing at least one auto-contouring performance measure in relation to each training atlas t within at least a subset of a set of T training atlases;

for each pair combination of candidate atlas n and training atlas t for which an auto-contouring performance measure has been computed, deriving a score based at least partly on the respective auto-contouring performance measure;

for each candidate atlas n within the set of N candidate atlases, aggregating the derived scores for each pair combination for which an auto-contouring performance measure has been computed for that candidate atlas n; and

selecting at least one subset of M candidate atlases from the set of N candidate atlases based at least partly on the most favourable aggregated scores therefor.

15. The non-transitory computer program product of claim 14 , wherein the non-transitory computer program product comprises at least one from a group including: a hard disk, a CD-ROM, an optical storage device, a magnetic storage device, a Read Only Memory, ROM, a Programmable Read Only Memory, PROM, an Erasable Programmable Read Only Memory, EPROM, an Electrically Erasable Programmable Read Only Memory, EEPROM, and a Flash memory.