Modeling plan-related clinical complications using machine learning tools in a multiplan IMRT framework

Abstract

Purpose: To predict organ-at-risk (OAR) complications as a function of dose-volume (DV) constraint settings without explicit plan computation in a multiplan intensity-modulated radiotherapy (IMRT) framework.

Methods and materials: Several plans were generated by varying the DV constraints (input features) on the OARs (multiplan framework), and the DV levels achieved by the OARs in the plans (plan properties) were modeled as a function of the imposed DV constraint settings. OAR complications were then predicted for each of the plans by using the imposed DV constraints alone (features) or in combination with modeled DV levels (plan properties) as input to machine learning (ML) algorithms. These ML approaches were used to model two OAR complications after head-and-neck and prostate IMRT: xerostomia, and Grade 2 rectal bleeding. Two-fold cross-validation was used for model verification and mean errors are reported.

Results: Errors for modeling the achieved DV values as a function of constraint settings were 0-6%. In the head-and-neck case, the mean absolute prediction error of the saliva flow rate normalized to the pretreatment saliva flow rate was 0.42% with a 95% confidence interval of (0.41-0.43%). In the prostate case, an average prediction accuracy of 97.04% with a 95% confidence interval of (96.67-97.41%) was achieved for Grade 2 rectal bleeding complications.

Conclusions: ML can be used for predicting OAR complications during treatment planning allowing for alternative DV constraint settings to be assessed within the planning framework.

Figure 1

Axial slices of the (a) head-and-neck case showing the planning target volume (PTV), parotids, and spinal cord, and (b) prostate case showing the PTV, bladder, and rectum.

Figure 2

DVHs corresponding to full knowledge base (encompassing both training and testing datasets) of 125 head and neck plans for the (a) left parotid and (b) right parotid and (c) spinal cord.

Figure 3

DVHs corresponding to full knowledge base (encompassing both training and testing datasets) of 256 prostate plans for the (a) bladder and (b) bowel and (c) rectum.

Figure 4

Summary of Modeling Process Summary involving ML prediction of OAR complications using features and plan properties (if necessary).

Figure 5

Optimized Decision Tree Algorithm Schematic.

Figure 6

Comparison of the modeled plan dose as a function of the input constraint settings using quadratic modeling and the actual achieved dose for the left parotid at volume levels of (a) 24% (b) 45% and (c) 66%.

Figure 7

Comparison of the modeled plan dose as a function of the input constraint settings using quadratic modeling and the actual achieved dose for the right parotid at volume levels of (a) 24% (b) 45% and (c) 66%.

Figure 8

Comparison of the mean (obtained from the 2-fold cross-validation process) predicted saliva flow rate (normalized to the pre-treatment saliva flow rate) using equation 5 to the actual saliva flow rate (calculated using equation 1).

Figure 9

Decision Tree for Grade 2 Rectal Complication Classification – an example.

Figure 10

Prediction of saliva flow rate (expressed as a percentage of the pre-treatment saliva flow rate) as a function of the dose constraint settings for the three OARs: (a) fixed cord dose constraint, dose constraint ranges for LP and RP, (b) fixed RP dose constraint, dose constraint ranges for cord and LP and (c) fixed LP dose constraint, dose constraint ranges for cord and RP.

Figure 11

Prediction of Grade 2 rectal complications as a function of the dose constraint settings for the three OARs: (a) fixed bowel dose constraint, dose constraint ranges for bladder and rectum, (b) fixed bladder dose constraint, dose constraint ranges for rectum and bowel and (c) fixed rectum dose constraint, dose constraint ranges for bladder and bowel.