ESTRO 2024 - Abstract Book

S3697

Physics - Dose prediction, optimisation and applications of photon and electron planning

ESTRO 2024

To assess the advantages of the new biological optimization algorithm, four different tumour cases were selected: breast, head and neck, pancreas, and prostate cancer. Patient data was acquired from matRad and PLUNC (Tewell et al., 2004) libraries as well as the Cancer Imaging Archive (Clark et al., 2013). Plans obtained with inverse physical optimization reproduced as much as possible the criteria used in the clinics for plan approval. Plan comparison was made based on TCP, NTCP and P+, as well as common dosimetric measures. For physically optimized plans prescription doses were as follows: for breast cancer dose 50 Gy delivered over 25 fractions; for pancreatic cancer 59.4 Gy delivered in 33 fractions; for head and neck cancer 70 Gy were prescribed to the primary tumour while for surrounding lymph nodes the prescription dose was 56 Gy to be delivered in 31 fractions; and for prostate cancer 74.4 Gy to the primary tumour and 55.8 Gy to the lymph nodes delivered in 31 fractions. Beam directions were the same for both physical and biologically optimized plans. For breast cancer four beam directions were used (0°, 45°, 155°, 310°), while for the remaining cases equidistant beam directions were used: 9 for head and neck and 7 for pancreas and prostate.

Results:

In table 1 the total TCP, NTCP and P+ (TCP-NTCP) for each tumour case using physical and biological optimization, respectively, are presented. For all cases, a significant increase in P+ was obtained for biological driven treatment plans due to a small reduction in TCP but a significant reduction in NTCP. Additionally, biologically optimized plans presented a more heterogeneous dose distribution but also a higher overall dose in the target volume, i.e, in general D98% and D2% with biologically optimized plans were smaller and larger, respectively, when compared to physical optimized plans allowing this to significantly spare surrounding organs at risk while increasing the mean dose in the PTV. Common dose tolerance with biologically optimized plans were always accomplished.

Conclusion:

Radiobiological models are valuable tools that, when used as objective functions in treatment plan optimization, can produce plans with significantly better treatment outcomes than those produced with physical optimization. Additionally, the versatility of the biological models makes it relatively easy to introduce more detailed radiobiological processes. The work here presented opens the door to future work, namely the use of better optimization algorithms and improved models. An improvement that will be explored is the reduction of the optimization time by running the current algorithm in a GPU and exploring parallel computation.

Keywords: Biological optimization, Simulated annealing