Home
The University of Adelaide
| The University of Adelaide |
Home |
Departments |
Search
|
 |
||
 |
 |
  |
 |
||||||||||||||||
|
|
Computer simulations of lymph node metastasis for optimizing the pathologic examination of sentinel lymph nodes in patients with breast carcinomaFarshid, G., Pradhan, M., Kollias, J., Gill, P.G. IntroductionThe detection of nodal metastases is highly dependent on the sampling methods used to look for them[ 1] . If sentinel node (SN) biopsy is to be used as a viable alternative to axillary dissection for staging breast cancer, histologic protocols used for detecting metastases in these nodes must have definable and acceptably low false negative rates (FNR). We also need a rational basis for comparing the reported experience of various centers involved in SN research. In this study we have devised a computer-generated model of a SN (a virtual SN) and have used it to compare the likely efficacy and cost of various sampling strategies in detecting metastases of specified sizes. This model may be used for the purpose of comparing the relative merits of various policies so that pathologists may make informed decisions about the protocols they chose to adopt in their laboratories. These simulations will also facilitate comparisons between the reported results of various centers involved in SN research. MethodsWe retrospectively examined 112 SNs removed from 89 breast cancer patients treated at our institution and recorded the node size and the number, size and position of any nodal metastases. This histologic data was used to devise a mathematical model (figure 1) of a SN with Matlab modeling software[ 2] .
Figure 1: model of sentinel node with metastasis We then simulated examination of this computer-generated (virtual) node according to a number of gross and histologic sampling protocols and for each protocol assessed the probability of detecting micrometastases of specified sizes. We used published costing figures to estimate the cost of the policies. ResultsAssuming zero observer error, bisection of a SN and examination of a standard 4 micron (µ) histologic section has a 64% probability of detecting a 2mm metastasis. This figure is <1% for a 50µ metastasis. Fine slicing the SN at the time of grossing improves the detection rate (see figure 2). This improvement is maintained for a range of metastasis sizes, is in the order of 2-4 fold and because all slices can usually be submitted in the one block, is achieved for a small increase in cost.
Figure 2: Effect of grossing strategy on detection of metastases. The examination of step-sections (levels) also improves the FNR. Our simulations highlight the importance of exhausting the tissue block with step sections . Protocols that do not exhaust the block have the attraction of fixed cost of processing but their performance deteriorates with increasing node size. By contrast, policies that advocate exhausting the block have a constant expected efficacy for detecting metastases of specified sizes but their cost increases with the size of the SN because all slices of larger nodes cannot be submitted in the one tissue block. The minimum size of metastasis one wishes to detect dictates the intervals between sections. For each size metastasis the improvement in FNR reaches a plateau after a set number of sections. Direct comparison of 6 sectioning strategies currently in use by pathology laboratories showed the chances of detecting a 500µ metastasis ranged from 13 % to 75%. Four of the 6 protocols had a less than 25% chance of detecting metastases of this size. Their costs ranged from $28-$50. Cost was not a good discriminator as some policies were more efficient than others were (see figure 3). We have also used genetic algorithms [ 3 4] to search for optimal policies given cost constraints.
Figure 3: A comparison of SN examination policies References
|
|
|||||||||||||
