ESTRO 2024 - Abstract Book

S4512

Physics - Machine learning models and clinical applications

ESTRO 2024

Areas under the ROC curve (AUCs) of toxicity predictions are presented in Figure 2. It can be observed that predictions were significantly better when using the mean doses of the identified anatomical subregions (median AUCs of 0.78 and 0.74, for the ALC and ANC-based endpoints, respectively).

Figure 2. AUCs of haematological toxicity predictions . Two haematological toxicity endpoints, based on (a) the absolute lymphocyte count and (b) the absolute neutrophil count, were predicted using the mean doses of the ilium, the lumbar spine, the pelvic bone, the bone (ilium + lumbar spine + pelvic bone) and the identified anatomical subregion.

Conclusion:

Voxel-based analysis allowed to identify anatomical subregions where the planned dose distribution is associated with haematological toxicity. Their mean doses were significantly more predictive than those of the whole bone regions. These results pave the way for personalised irradiation planning sparing these subregions.

Keywords: Prostate, voxel-wise analysis, toxicity

References:

This work was funded by ERA PerMed.