Isocenter position is a practical radiation concern in a few ways. This will be off the top of my head, and there may be other scenarios which I'm not considering. There are also different styles of doing this, and this just represents my particular thought process. I'll start basic and build up. I'm going to ignore tomotherapy and cyberknife for now unless someone responds with questions about how this applies to those.
For those reading who have no concept of what the isocenter is: the physical isocenter is the point at which conventional linear accelerators rotate around. A typical radiation source in the machine head rotates in a 360 degree arc around a fixed point in space. It looks something like this:
(coronal view adapted from: http://www.jacmp.org/index.php/jacmp/article/view/4309/3073)
(MSPaint sagittal oblique view adapted from wikipedia)
When a patient undergoes simulation, a point is chosen within the patient called isocenter that corresponds to the treatment machine's physical isocenter. From a physics and dosimetry standpoint, generally the best place for an isocenter is often in the center of the target (PTV, more on this later). In a modern CT simulation, the patient is lying on the bed/table, supposedly holding perfectly still from the time the CT scan is performed, to the time the isocenter is set and the patient is marked. The marks (temporary or permanent, e.g. tattoos) represent the surface positions on the patient to which the isocenter point corresponds deep within the patient (typically in the target). The marks are usually 2 lateral points and a single anterior or posterior point depending on how the patient is lying (i.e. if lying supine, a mark goes on their anterior body surface at the sup-inf location corresponding to isocenter).
There are a few practical problems here. The first is that an attending may not be available for the simulation. Thus the responsibility for setting isocenter often falls in the therapists. My understanding is that this is a billing gray area.
In any case, let's assume you have a gross target and want to set isocenter in the middle of it for now. If you need to contour the tumor volume, CTV expansions/nodes at risk, PTV, etc, there may not be sufficient time from time of scanning to the set the isocenter perfectly inside the PTV. This may take hours for complicated cases or even if it only takes minutes the patient may be in pain and may not tolerate lying still for long at all. Thus, fast contours of the tumor volume can be chosen on every other or third slice, interpolated, and the isocenter could be set in the middle of that, knowing that you plan to modify the contours later. Or, you can "eyeball" the tumor on the 3D scan and just pick a point that seems to correlate to the center of it. In a palliative or gross tumor (like a pre-op sarcoma) case, I often find the top slice of tumor extent, the bottom slice of tumor extent, and then do a little simple math to find the middle of those two slices and then place the isocenter in the middle of the tumor at that slice. In a situation where there is a large CTV or only a CTV, at simulation I often set the expected field borders and then place isocenter in the middle of that field.
In the days before 3D planning, the isocenter could be determined clincially based on landmarks or based on 2D films +/- bony landmarks and set to a depth. In the 2D or clinial setup days, field arrangements, matching fields, surface/bony anatomy to determine organs at risk, blocking factors like shapes and calculations for custom blocks and wedges, were highly important before 3D planning and MLCs but are easily adjusted today. E.g. You didn't want to have a custom block made and then have to re-do it because you screwed up where the tumor really was. Or, you didn't want to set your depth too shallow and underdose the target. Or you didn't want to have your field going through a critical structure that would seem obvious now on a 3D scan. Thus, I've noticed that some older attendings pay particular attention to how the isocenter is selected compared to some younger attendings.
There are some cases even in the 3D era where isocenter position remains of critical importance for field matching or divergence. For example, in the monoisocentric breast technique, the isocenter is set at level of the inferior clavicular head (for pictures of this see:
http://clinicalradiotherapy.weebly.com/breast.html). Since there is no field divergence at isocenter, this allows you to match multiple fields at that point for treatment. Another commonly tested treatment where all of these concepts of gantry angles, collimator angles, couch kicks, skin gaps, feathering of junctions, are tested is in the photon craniospinal setup (without tomotherapy).
Even eliminating these special cases where isocenter needs to have specific positions, we have to keep in mind that the collimator size is limited. Conventional linear accelerators generally have a size on the order of 40 x 40 cm. From the isocenter, that's 20cm in each direction. If you have a large field, you may not be able to encompass the entire target unless the isocenter is in the center of the target. The best way to evaluate this is on a "beam's eye view" which looks like the following when rendered using reconstructed bony anatomy from the simulation CT:
http://synapse.koreamed.org/DOIx.php?id=10.3857/roj.2013.31.4.252&vmode=PUBREADER
Now on this pelvis field, you can see the isocenter position (where the two lines labeled X and Y intersect) is not in the exact center of the field. That's ok. The small hash marks are 1cm, the large hash marks are 5cm. You can see that on the A-P view (A), the field is about 11.5cm in Y1. You can imagine some point at which if the isocenter was set too high, the field would not be able to open far enough (past 20cm) to cover the target.
So what if you do set the isocenter in the wrong place or decide later it needs to be changed? Generally you have two options. You can place new marks on the patient. That's easy with temporary markings, but patients don't like getting more tattoos than they have to. So instead, you can instruct the therapists to perform daily isocenter shifts. That is, line up to the marks on the patient, and then move the table so many centimeters in some direction to the new treatment isocenter. The problem with either of these approaches is that there is margin for error. If the therapists shift in the wrong direction one day, you will miss completely. You may not even know it happened unless you're doing daily image guidance.
So that is what happens if you're grossly off of the right position. There's a second, more subtle and more recent issue I can't find a good picture to describe. That is that many accelerators now use MLCs that have multiple widths depending on distance from isocenter. In the pelvis field above, each of those little white boxes on the sides represent the position of a single MLC "leaf" which move in and out (left-right on the images). The width (in the sup-inf axis) is determined based on how the MLC is manufactured. For example, the "High Definition" Varian MLC has 8cm of 2.5mm width MLCs, then 5mm width MLCs, but there are many different MLC products of different sizes and widths. The finer the MLCs, the easier it is to shape your field to the target and spare critical structures. Thus, it is often desirable to have finer width MLCs on your target, and not the thicker ones.
Another practical concern is that the beam's eye view can be a digitally reconstructed radiograph (DRR). DRRs are pictures made by the planning system. They use the attenuation values from the simulation CT to reconstruct what a 2D x-ray film would look like for various positions of the imager (i.e. for the image A, above, that is that the x-ray source is above the patient, and for B that is that the x-ray source is lateral to the patient). So for positioning that pelvis in the images above for treatment, I would compare the DRRs to actual A-P and lateral x-rays of the patient in treatment position that are obtained on the first day of treatment and weekly. Now depending on the bony landmarks of the patient, if your target is very much in soft tissue or is in a place that has complicated anatomy (like the head and neck for example), it may make more sense to set the isocenter somewhere where it is easy to align. So some attendings pick an isocenter in a standard position for that site. For example, for many head and neck tumors, that spot may be the anterior-inferior point of the C2 vertebral body. That makes the setup images much easier to evaluate.
Hope that all makes sense. I'm out of time for now.
