GB2335583A - Controlling delivery of radiotherapy - Google Patents

Abstract

A system for discrete beam intensity modulation uses a multi-leaf collimator to determine irradiation from a linear accelerator. The leaves of the collimator are adjusted during irradiation (Fig.2a), and the diaphragms or jaws of a further collimator are advanced beyond the collimator leaves (Fig.2b) when necessary so as to prevent over-dosing while maintaining at least minimum leaf separation. An algorithm is used to control the movement of the collimators.

Description

2335583 CONTROLLING DELIVERY OF RADIOTHERAPY The present invention relates

to a method of controlling the delivery of radiotherapy. It uses a multileaf collimator to deliver radiation doses which vary over the treatment area.

A number of authors have described how intensity-modulated fields for conformal radiotherapy can be implemented by either dynamic multileaf collimation (for example, K511man et al (1988), Convery and Rosenbloom (1992), Spirou and Chui (1994), Stein et al (1994), Svensson et al (1994), Yu et al (1995), van Santvoort and HOmen (1996) Hill et al (1997)) or by multiple static fields (for example, Galvin et al (1993), Bortfield et al (1994).

US 5663999 discloses a method in which the treatment area is divided into a plurality of sections which are treated individually.

It will be clear that the quickest way to achieve a variable dose treatment would be to irradiate various parts of the treatment area simultaneously. The system of US 5663999 is unable to do so, although it notes that treatment time is a factor which will ideally be reduced. However, it has hitherto been difficult or impossible to provide an algorithm able to deliver any treatment profile, for the simple reason that the multi-leaf collimators available at present must maintain a minimum leaf separation of (for example) 1 centimetre. As a dynamic treatment pattern evolves, it may be necessary to blank off parts of the treatment area in a way which calls 2 for leaves to touch. Thus, some treatment profiles cannot be delivered, preventing widespread application of this technique.

The present invention provides a method of delivering a radiotherapy treatment using a linear accelerator, the linear accelerator comprising a source of radiation, the output of which is limited by a multi-leaf collimator and a further collimator comprising at least two diaphragms; the method comprising the steps of: notionally dividing the treatment area into an array of cells distributed along fines parallel to the movement directions of the leaves; assigning an intended dose to each cel. l; during irradiation, adjusting the position of the leaves so as to provide the intended dose to each cell, wherein during irradiation, the diaphragms are advanced beyond one or more leaves when necessary to maintain leaf separation and prevent overdosing.

Thus, according to the present invention, the diaphragms are normally maintained behind the outermost leaf, but are on occasions advanced beyond one or more leaves so that they provide the primary shield against dosing. This effectively permits one or more leaves to be withdrawn, allowing an opposing leaf to be advanced so as to limit dosage without infringing minimal leaf separation.

It is good practice to keep the diaphragm immediately behind or close behind the outermost leaves. This is because multi-leaf collimators are somewhat leaky and therefore the diaphragm will limit some leakage.

In this application, it is assumed that leaves on the left of the field are extended as leaves on the right are withdrawn. All directional detail is to be interpreted in this context. However, it will clearly be understood that the invention can be applied to a reversed situation.

3 An embodiment of the invention will now be described by way of example, with reference to the accompanying figures, in which; Figures 1 a and 1 b show a typical discretely-intensity modulated beam; Figure 2a and 2b illustrates the minimum leaf separation constraint; Figure 3 illustrates the control point formulation; Figure 4 shows a schematic flow diagram; and Figure 5 illustrates the control point sequence to achieve the beam of Figures 1 a and lb.

A discretely-intensity-modulated beam is here taken to mean one which is divided into a number of smaller beam elements. Within each beam element the intensity is constant but each element may be at a different intensity level. An example is shown in Figure 1. This is for the posterior oblique field of a prostate plan with 10% intensity stratification and 'I cm x 1 cm beam elements at the isocentre plane, planned with Corvus planning system (example courtesy of NOMOS). Figure 'I a shows the example in grey-scales and Figure 1 b shows the intensity matrix. Beam elements hereafter referred to as "bixeis" (by analogy with image pixels) typically measure 1 cm x 1 cm at the isocentre. Discrete beam intensity modulation is assumed by some inverse treatment planning systems.

Implementing discrete beam intensity modulation (BIM) by continuous dynamic multileaf collimation (DIVILC) poses some unique problems. In particular, abrupt intensity changes are required between beam elements along the direction of leaf travel. Hill et al (1997) have previously shown that this can be achieved by combining the ramp down of intensity delivered across a bixel as a leading leaf moves across it at constant speed with the 4 corresponding ramp up delivered as the trailing leaf passes it with the same speed. A similar use of overlapping ramped intensity distributions was used by Davy and colleagues on the Tracking Cobalt Unit at the Royal Free Hospital, London (see for example Davy (1985)).

The algorithm presented by Hill et a] (1997) is an extension of that used by Bortf ield et al (1994) for generating discrete intensity-modulated beams by summing a series of static IVILC fields. In these algorithms, the intensity profile along each leaf pair is examined independently. For each profile, rising and failing edges of the intensity pattern are paired to give a series of static fields, which can be ordered to give a unidirectional "sweep" of the leaves across the field. In the dynamic implementation, Hill et al (1997) transform this multiple static field sequence into a single dynamic sequence during which leaves can be either stationary or moving at the maximum allowed speed and the previously described ramp down and ramp up of intensity across bixels then overlap to give uniform bixel intensities. No consideration of tongue-and-groove artefacts (van Santvoort and Heijmen (1 99C is made in the algorithm.

The above dynamic scheme requires that the leaves can close to zero separation at the start and end of the sequence and also that a trailing leaf of one leaf pair can overlap an adjacent leading leaf on the opposite leaf bank. This is not possible on all multileaf collimators and indeed may not always be desirable due to the associated risk of mechanical collisions and the leakage through the touching rounded leaf ends used on some MLC designs.

For the multileaf collimator (MLC) produced by the applicants, the minimum allowed leaf separation in dynamic mode is set to 1 cm with this restriction also applying to opposite but adjacent leaves so that no leaf overlap can occur. Referring to Figure 2(a), the most leading left leaf must not only maintain the necessary minimum separation with the opposite right leaf but also with the right leaves above and below. The same minimum separation distance applies between the backup diaphragms. Projected to the isocentre, the MLC leaf width is 1.Ocm and their range of travel is 20cm from the beam axis to 12.5cm over midline. The MI-C's backup diaphragms have the same range of travel, while the collimator pair perpendicular to this can travel from 20cm from the beam axis up to midline. Details of the Elekta MLC and its characterisation for static field use has been given previously by Jordan and Williams (1994).

Dynamic control of the IVILC leaves and diaphragms is specified in the algorithm described herein through the use of the "control point" formalism employed by the Elekta dynamic MLC control system. A control point specifies the machine configuration (field size and shape, gantry angle, etc.) at a particular percentage of the set beam monitor units (MU). By defining a sequence of control points associated with different percentage monitor units a dynamic delivery is defined, for example as shown in Figure 3. In the Elekta control system, the desired value of any parameter (e.g. an MLC leaf position) at any particular percentage monitor unit value is found by linear interpolation between the control points immediately before and after this percentage MU. By using percentage MU rather than actual beam MU, dynamic beam prescriptions are readily scaled with the set MU for the field. Control points do not need to be evenly spaced.

Any beam must be defined by at least two control points, an initial control point (control point 0) that defines the starting configuration and one defining the stop configuration. Control points defining intermediate configurations can be added.

In addition to dynamic beam collimation, this formalism can be used to define static and move-only beams or beam segments. A static beam segment is defined by two consecutive control points having the same parameter values but different percentage MU, whilst a move-only step can 6 be implemented by two control points with the same percentage MU but different parameter values. The beam is cut off during such move-only steps by inhibiting the radiation output from the linear accelerator. Combinations of static, dynamic and move-only segments may be mixed within a single beam prescription.

Note that although this control point formalism specifically supports one particular MLC control system, the same ideas can be used on other manufacturers' MLCs. Note also that, as will be shown below, the results generated by the algorithm described in this paper are also equally suitable for implementation via multiple static MLC fields.

Generation of discrete intensity-modulated beams by dynamic collimation Ignoring for the moment restrictions on the allowed minimum leaf or collimator separation, the basic idea used to generate a discrete intensitymodulated beam by dynamic collimation can be illustrated by considering one isolated beam element. As the leading leaf or collimator jaw moves at constant speed to expose the element, a ramp down in intensity is delivered across the element. The element is then uniformly irradiated until the trailing leaf or jaw, moving at the same speed as the leading leaf, occludes the element and generates the necessary ramp- up in intensity needed to give a sharp, discrete primary exposure. This is basically the technique used by Hill et al (1997). Note that at the start and end of a row of these elements the leaf/jaw separation needs to be zero in order to provide a sharply-defined beam edge.

Generation of discrete intensity-modula ted beams under minimum separation constraints The process described above requires that the leaves or collimators can close in to zero separation at the start and end of the exposure. As has 7 been noted, this is not possible on all IVILCs. However, according to the invention, it can by simulated by combining the motion of the MLC leaves and with that of the backup diaphragms (collimator jaws) in the dynamic beam prescription. In this case the positions of the leaves and backup diaphragms are offset such that the required gap between the MLC leaves is "hidden" under the backup diaphragm. This is illustrated at Figure 2(b).

Central blocking Central blocking - that is, blocking a region inside the main field boundary - can also be achieved during a single dynamic exposure without having to close the leaves together, in which case the necessary leaf gap is again hidden under a backup diaphragm in the region to be blocked, as in Figure 2(b).

Implications of minimum leaf separation constraints for two-dimensional beam intensity modulation Minimum leaf separation constraints, and in particular constraints on the allowed positioning of adjacent leaf pairs (Figure 2(a)), impose restrictions on the way in which a beam's modulation can be delivered. Specifically, when generating a two-dimensional (2D) intensity modulation within a beam a trailing leaf of one leaf pair may need to be moved forwards to shield a region so as to prevent it from being overexposed. Maintenance of the minimum leaf separation distance can then push a neighbouring leaf pair's leading leaf forwards, potentially leading to an overexposure here if only the MLC leaves are being used to generate the modulation. If, however, the backup diaphragm can be used as part of the dynamic collimation process then the leading leaf bank's backup diaphragm can be used to provide the necessary shielding, so allowing the leading leaf to move forwards whilst avoiding any overexposure. Alternatively, it may be possible to "slow down" the generation of the intensity distribution under the original 8 leaf pair so that the trailing leaf moves forward at a later point in the delivery sequence when it will then not give rise to a violation of the minimum leaf separation constraint with the neighbouring leaf pair. This is discussed in more detail below.

in general, then, under these minimum leaf separation constraints it is found that the derivation of the leaf motion for one pair of leaves cannot be considered in isolation from those of its neighbours because of the influence of "knock-on" effects such as that described above. Rather, it is necessary to consider together all involved ieafs pairs and the profiles they are separately generating at all stages of the leaf motion generation. This then leads to a different approach to deriving MLC and diaphragm motion in which the derivation is made progressively controlpoint-by-control-point. In this way full account can be made of restrictions on leaf positioning (including adjacent leaf positioning) during the derivation and, as will be shown below, tongue-and-groove artefacts are then also easily removed. This approach is used in the algorithm described below.

---- 000 ---- The Bixel Beam Intensity Modulation algorithm (Bixel BIM) is an iterative scheme to derive MLC leaf and backup diaphragm motion to deliver discrete beam-element (bixel) intensity modulated fields subject to restrictions on the allowed leaf separation. The algorithm progressively derives the motion of all involved leaf pairs and the MI-C's backup diaphragms co ntro 1-poi nt-by-co ntroi- point as they track across the field rather than by considering each leaf pair's derivation separately, and includes a means of feeding back information on minimum leaf gap violations in order to derive deliverable dynamic prescriptions. It is assumed in what follows, without any loss of generality, that the leaves and diaphragms move from left to right across the field.

9 In deciding whether or not a bixel can be irradiated at any given control point two logical arrays (truth tables) are used within the algorithm. The first, shield, is defined for each bixel at each control point. It is set to "true" if a bixel must be shielded at the given control point; shielding may be by left or right MILC leaf or backup diaphragm. The second array, right_diaphragm-shield, is defined for each column in the field for each control point. If this is set to true for a given control point, then this entire column of bixels must be shielded by the right backup diaphragm (collimator jaw) at this control point. The status of these arrays is partly defined by the required intensity distribution itself (for example, any zero-intensity bixel must always be shielded), but are also continuously modified as the algorithm runs to correct for or prevent leaf collisions or minimum gap violations, to avoid overexposing regions of the field and to remove any tongue-and-groove artefacts within the derived motion. Note that, by definition, if right_diaphragm-shield is "true" for a column at any given control point, then so must shield for all bixels within that column.

The algorithm deals in integer units of bixel intensity, where one unit is equivalent to the intensity delivered in the time it takes for a leaf or -diaphragm to travel the width of one bixel. The absolute magnitude of this intensity therefore depends on the width of the bixel, the maximum available leaf speed and on the dose rate used to deliver the field. MLC leaves and diaphragms are allowed to travel a distance of one bixel-width per control point and the minimum leaf separation is specified within the algorithm in units of bixel-widths. The mapping of a desired intensity distribution into bixel intensity units, and the control of the delivery resolution, is discussed in detail below. The minimum separation distance is expressed in units of bixel widths.

A schematic flow diagram of the algorithm is shown in Figure 4. A description of each of the main steps is given below.

Initialise leaf and diaphragm settings (control point 0) At the start of the irradiation, the left leaves are set at the furthest left boundary of the field and the right leaves lie a distance equal to the minimum separation (in units of bixel-widths) into the field. The backup diaphragms are offset from these by a distance equal to this minimum bixelunit separation so that the right diaphragm is at the left boundary and the left diaphragm is outside of the field.

Loop over control points Starting from the first control point, use the current shield and right- diaphragm-shield settings to derive motion control-point-by-controlpoint as the MI-C leaves and diaphragms track across the field. Thus, the steps below are carried out at each control print.

Set left backup diaphragm position If any bixels in the column to the right of the current left backup diaphragm position requires at least one more control point to finish off its irradiation (found by checking the cumulative intensity so far delivered there), then the diaphragm must remain where it was at the previous control point. Otherwise, move it one bixel-width to the right.

Set left MLC leaf positions A left (trailing) IVILC leaf is only moved forward when the bixel after its current position is to be occluded at this control point (found by checking the cumulative intensity so far delivered there). if at least one more control point is needed to complete its irradiation, it maintains its position from the previous control point.

Set right MLC leaf positions For each right MLC leaf, if the bixel that would be exposed by moving this leaf one bixel-width to the right is required to be shielded at this control point (i.e. shield is "true" for this bixel at this control point), right_diaphragir_shield is "false" and the left backup diaphragm has not been set to shield it at this control point, then it is necessary to keep the right leaf where it was at the previous control point. Otherwise, the leaf position for the current control point is set to one bixel-width further to the right than that at the last control point.

Set right backup diaphragm position In general, the right backup diaphragm position at each control point is set to the position of the furthest-most right MLC leaf. However, it may be set less far into the field if the right_diaphragf77_shield logical array is set to "true" for this column and control point, so that the diaphragm is being used to proving the necessary shielding here. Likewise, it may be necessary to hold the right diaphragm at the right field boundary to prevent overdosage outside of the field.

Check for collisions or minimum separation violations Checks are made at this stage to ensure that the leaf and diaphragm positions, as set by the previous steps, do not result in violations of the minimum separation criteria. Checks must be made separately of the diaphragm separation and of the separation of the MLC leaves.

The backup diaphragm separation is checked first since their positioning can affect how MLC leaf separation violations are resolved. If it is less than the minimum allowed, then the right diaphragm is moved forward one bixel width unless it is acting to shield the column it is currently 12 above (as indicated by the right-diaphragm_shield logical array), in which case the left diaphragm must instead be set back one bixel width. If the left diaphragm has been moved back, it is then necessary to further check if this will then overexpose anything the diaphragm had been shielding. If this is found to be the case then right_diaphragT_shield is set to "true" for this column up until the current control point and the motion derivation restarted taking this new information into account.

Once the backup diaphragm positions have been checked and if necessary corrected, the MLC leaf positions can then be assessed. As noted previously (section 2), for designs such as the Elekta IVILC the minimum separation requirement applies not only to each individual leaf pair but also to the positions of the neighbouring leaves, each of which must therefore be checked.

According to this example, a leaf separation check is made for each right leaf in turn. This check could of course be carried out for each left leaf.

If left and right leaves are within the minimum separation distance, a choice has to be made as to whether to move the left or right leaf to correct for this.

If the bixel the right leaf end is currently above does not need to be shielded at this control point or if it does need to be shielded and right-diaphragm-shield is also set, then the right leaf can be moved forwards.

If these conditions do not apply then a check is made to see whether it would be possible to prevent this constraint violation occurring by forcing the left leaf involved to arrive at this point later in the prescription (this can be done by inhibiting the motion of the associated right leaf). To do this, a 13 recursive check back through each column is made to find the first control point at which the bixel under the left leaf is exposed and assessing whether having the right leaf held back to shield this position would violate the minimum leaf separation requirement at that control point. If inhibiting the left leaf in this way is possible, the logical array shield is modified accordingly and the motion re-derived taking this into account. (For efficiency, the derivation may continue from the control point before the first change to shield).

If inhibiting the left leafs progress in this way is not possible, then the right leaf must be moved forwards and the right diaphragm forced to shield the bixel (and hence the entire column) by setting righ_diaphragmshield to be true up until this control point and re-deriving the motion. As this results in a less efficient delivery scheme - the diaphragm is shielding everything beyond this point - this option is only used when inhibiting a right leaf if motion (above) is not possible.

Check that no right leaf has gone too far under the right diaphragm After the minimum separation checks (above) and adjustments, a check is made to see if any right leaves under the right backup diaphragm are unnecessarily far under it. If they are and their settings for this control point can be adjusted without violating the minimum gap criteria, their positions are redefined. This step is done to minimise leakage through the backup diaphragm.

Check that no bikel is overexposed Following adjustments required by minimum separation conditions, a check is made that no bixel is overexposed as a result of the current leaf and diaphragm settings - this can occur, for example, if a right leaf has been moved forward to prevent violation of minimum separation conditions. If 14 any bixel is overexposed, the shield logical array is set to "true" for this bixel for all control points up to the current one and the motion re- derived taking this updated information into account. Again, for efficiency, the derivation may continue from the control point before the first change to shield.

Check cumulative intensity delivered A check is made to see if the correct cumulative intensities have been delivered to all bixels. If they have been, a check is optionally made of the intensities in the inter-leaf tongue-and-groove regions (below). If not, at least one more control point is needed and the loop over control points continues.

Check tongue-and-groove regions A check is also optionally made of the intensities in the inter-leaf tongue-and-groove regions as each control point is derived. If any underexposure is found, this can be corrected for by inhibiting the motion of the most leading leaf pair (van Santvoort and Heijmen (1996)). Within the present algorithm, this is achieved by finding the earliest control point at which the associated bixel is exposed and setting shield to "true" for this bixel for N control points from this control point on, where the dose delivered over these N control points corresponds to the tongue-and- groove underexposure found. The motion is then re-derived taking this updated information into account and as before, the derivation may continue f rom the control point before the first change to shield was made.

It is noted, however, that the maintenance of the minimum leaf separation and associated collision avoidance routines tends to synchronise the leaf motion so that underdosage in the interleaf region is then not a major change to the delivery scheme and often does not involve any extra control points (and hence extra MU) for the total delivery.

15.

Example application The result of applying the above algorithm to the example distribution in Figure 1 is given in Figure 5, where the IVILC configuration at each control point is shown. Regions in black are shielded by both an IVILC leaf and backup diaphragm, regions in dark grey by an IVILC leaf only and in light grey by a backup diaphragm only. The bixel (beam element) dimensions are 1 cm x 1 cm and a minimum leaf/diaphragm separation of 1 cm specified. The closed configurations at the start and end of the delivery, required for a sharp field edge, can be clearly seen, as can the use of the backup diaphragms to provide shielding when the minimum leaf separation constraints limit leaf positioning. Note also that, as described above, leaves and diaphragms are restricted to move by only one bixel width per control point.

The above algorithm has been implemented to deliver discrete beam intensity modulation on Philips/Elekta SL25 with MLC. A schematic diagram of this implementation is given in Figure 6 and the main features discussed below.

Scaling the input intensity distribution The input intensity distribution (which may of course be arbitrarily normalised) is scaled according to the "nominal dose" to be delivered by the field. This is done to control the resolution of the delivered intensity distribution and to ensure that the dynamic beam prescription, when delivered, will not require leaf speeds in excess of those the MI-C is capable of (see below). The result can then be treated as a monitor unit distribution, specifying the MU each element in the beam should receive. Note, however, that the exact value used for the nominal dose is not critical (it is only used as a guide) since the beam prescription file is based on control points at percentage monitor unit points and is therefore scalable with set monitor

16 units.

Open field component delivery

In cases where the field is not central ly- blocked, an option exists within the implementation to deliver some or all of the base intensity level of the field by a single static field component. This would then allow portal imaging for positional verification of the patient as for conventional radiotherapy and could also allow for much of the field intensity to be delivered by a relatively large static field for which the dosimetry is more familiar and for which the effects of, for example, patient movement and leaf set-up errors are less critical. When this option is selected, the first two control points specify the static field (with the appropriate percentage MU). This is then followed by a moveoniy step during which the beam is off and the IVILC leaves move to the left field boundary, after which the dynamic modulation is delivered in the subsequent control point steps.

Controlling the resolution of the delivered intensity distribution The "basic" intensity resolution for delivery within this scheme is governed by the intensity delivered during the time it takes a leaf or diaphragm to cross the width of a beam element (bixel). This depends on the maximumallowed leaf speed, the dose rate at which the field is to be delivered and the element width. For example, for a 1cm-wide beam element, treating at 400MU/min with a maximum leaf speed of 1cm/sec results in a basic resolution for delivery of 6.67MU, i.e. 3.3MU. The monitor unit distribution (i.e. the scaled intensity distribution) is then rounded to the nearest MU/bixel to create a bixel-intensity-unit array for use internally within the algorithm.

This resolution of delivery can, however, easily be improved by simply subdividing as necessary the beam elements in the original distribution along 17 the direction of leaf travel before rounding to give the bixel-intensity- unit array. In the example above, if the 1 cm-wide elements were subdivided into 3.3mm elements, the resulting delivery resolution would be improved to 1. 1 MU. The penalty paid for this is an increased number of control points required to define the dynamic prescription. It does not, however, affect the IVILC delivery efficiency, i.e. the number of MU required to deliver the field.

In the algorithm described above, only the primary intensity has been considered - contributions from the finite transmission through the leaves and diaphragms, head scattered radiation and the variation of the primary intensity away from the central axis have not been included. This was intentionally done to provide a portable algorithm that was based on geometric concerns only and which was therefore independent of machinetomachine variations and also of methods of calculating the above quantities. Instead, these additional factors can be included by embedding the algorithm within an iterative loop, such as the following (Convery and Webb (1 997W 1 Calculate the leaf and diaphragm motion using the above algorithm (geometry only).

2. Use this leaf motion to calculate the total intensity or dose under reference conditions, as required. If the difference between the calculated and desired values is greater than a set tolerance (or if continued iteration provides no improvement) then exit the iterative loop, otherwise proceed to step 3.

3. Calculate a modified input intensity distribution based on the difference between the calculated and desired values.

4. Use this modified intensity distribution as the input to the algorithm - repeat from step 1.

It is important to note that by separating dosimetric aspects from the actual derivation of the MLC motion the algorithm becomes independent of 18 the quantity we are actually concerned with modulating, which may for example variously be the total intensity of fluence "in-air" at the isocentre plane, the dose at peak depth at the isocentre plane, the dose at some reference depth under specified reference conditions, or some other quantity. The algorithm's application is also independent of the method used to calculate these quantities, so that it may therefore be used within different planning systems or software modules without modification, so ensuring portability and wider applicability.

Although the algorithm described was originally developed to generate dynamic collimation schemes for the delivery of fields with discrete beam intensity modulation, it can be seen from Figure 5 that the output can equally well be interpreted as a sequence of static M LC fields (defined by the control points configurations excluding those at the start and end of the motion). Since the original design was for a dynamic implementation, the beam-off time involved when stepping between each field configuration would be small. The only change required to the algorithm would be to write out each derived step as a static control point pair to be followed by a move-only step to the next field configuration. This may be necessary, for example, where full dynamic control of the MLC is not available.

Multiple static field delivery may also be preferred since this is a smaller step away from conventional radiotherapy delivery. Dosimetry of these beams is easier, as is their QA and verification, and they may therefore be more readily accepted clinically.

Although the techniques described herein (derivation of motion controlpoint-by-control-point, leaf collision correction/prevention routines, use of the backup diaphragms as part of the dynamic collimation process) were developed within the context of developing dynamic delivery schemes for discretemodulated beams, they are also applicable to generating continuous (i.e. non-discrete) modulations.

19 Publications referred to in the above are as follows:

Bortfield T R, Kahler D L, Waldron T L and Boyer A L 1994 X-ray field compensation with multileaf compensators Int J. Radlat. Oncol. Biol. Phys 28723-730

Convery D J and Rosenbloom M E 1992 The generation of intensitymodulated fields for conformal radiotherapy by dynamic collimation Phys. Med. Biol 37 1359-1374

Convery D J and Webb S 1997 Calculation of the distribution of headscattered radiation in dynamically-collimated IVILC fields Proc 12th Int Conf Computers in Radiotherapy, Utah, May 1997 (Ed. LeavittDD andStarkschall G), pp 350-353

Das 1 J, Kase K R and Tello V M 1991 Dosimetric accuracy at low monitor unit settings Brit. J Radiol. 64 808-811 Davy T J 1985 Physical aspects of conformation therapy using co mputer-co ntro lied tracking units Progress in Medical Physics (Ed.. Orton C G), VOL 2 350-353 Hill R W, Curran B H, Strait J P and Carol M P 1997 Delivery of intensity modulated radiation therapy using computer controlled multileaf collimators with the CORVUS inverse treatment planning system Proc 12th Int Conf Computers in Radiotherapy, Utah, May 1997 (Ed. Leavitt DD and Starkschall G), pp 393-397 Jordan T J and Williams P C W 1994 The design and performance characteristics of a multileaf collimator Phys. Med. Biol. 39 231-251 K511man P, Lind B, Eki6f A and Brahme A 1988 Shaping of arbitrary dose distributions by dynamic multileaf collimation Phys. Med. Biol 33 1291 1300 Spirou S V and Chui C S 1994 Generation of arbitrary intensity profiles by dynamic jaws or multileaf collimators Med. Phys. 21 1031-1041 Stein J, Bortfield T, D6rschel B and Schlegel W 1994 Dynamic x-ray compensation for conformal radiotherapy by means of multi-leaf collimation Radiother. Oncol. 32 163-173

1 Svensson R, Kbilman P and Brahme A 1994 An analytical solution for the dynamic control of multileaf collimators Phys. Med. Biol 39 37-61 van Santvoort J P C and Heijmen B J M 1996 Dynamic multileaf collimation without "tongue-and-groove" underdosage effects Phys. Med. Biol. 41 20912105 Yu C X, Symons M J, Du M N, Martinez A A and Wong J W 1995 A method for implementing dynamic photon beam intensity modulation using independent jaws and a multi-leaf collimator Phys. Med. Biol. 40 769-787 21

Claims (4)

1. A method of delivering a radiotherapy treatment using a linear accelerator, the linear accelerator comprising a source of radiation, the output of which is limited by a multi-leaf collimator and a further collimator comprising at least two diaphragms; the method comprising the steps of: notionally dividing the treatment area into an array of cells distributed along lines parallel to the movement directions of the leaves; assigning an intended dose to each cell; during irradiation, adjusting the position of the leaves so as to provide the intended dose to each cell, wherein during irradiation, the diaphragms are advanced beyond one or more leaves when necessary to maintain leaf separation and prevent overdosing.
2. 2. A method according to claim 1 in which the diaphragms are normally maintained behind the outermost leaf, but are on occasions advanced beyond one or more leaves so that they provide the primary shield against dosing.
3. 3. A method according to claim 2 in which the diaphragm is otherwise maintained immediately behind or close behind the outermost leaves.
4. 4. A method substantially as described herein with reference to and/or as illustrated in the accompanying drawings.