The stability of imaging biomarkers in radiomics: a framework for evaluation.

This paper studies the sensitivity of a range of image texture parameters used in radiomics to: (i) the number of intensity levels, (ii) the method of quantisation to select the intensity levels and (iii) the use of an intensity threshold. 43 commonly used texture features were studied for the gross target volume outlined on the CT component of PET/CT scans of 50 patients with non-small cell lung carcinoma (NSCLC). All cases were quantised for all values between 4 and 128 intensity levels using four commonly used quantisation methods. All results were analysed with and without a threshold range of -200 HU to 300 HU. Cases were ranked for each texture feature and for all quantisation methods with the Spearman's rank correlation coefficient determined to evaluate stability. Results showed large fluctuations in ranking, particularly for low numbers of levels, differences between quantisation methods and with the use of a threshold, with values Spearman's Rank Correlation for many parameters below 0.2. Our results demonstrated the sensitivity of radiomics features to the parameters used during analysis and highlight the risk of low reproducibility comparing studies with slightly different parameters. In terms of the lung cancer CT datasets, this study supports the use of 128 intensity levels, the same uniform quantiser applied to all scans and thresholding of the data. It also supports several of the features recommended in the literature for such studies such as skewness and kurtosis. A recommended framework is presented for curation of the data analysis process to ensure stability of results.

Dose prescription complexity versus tumor control probability in biologically conformal radiotherapy.

The technical feasibility and potential benefits of voxel-based nonuniform dose prescriptions for biologically heterogeneous tumors have been widely demonstrated. In some cases, an "ideal" dose prescription has been generated by individualizing the dose to every voxel within the target, but often this voxel-based prescription has been discretized into a small number of compartments. The number of dose levels utilized and the methods used for prescribing doses and assigning tumor voxels to different dose compartments have varied significantly. The authors present an investigation into the relationship between the complexity of the dose prescription and the tumor control probability (TCP) for a number of these methods. The linear quadratic model of cell killing was used in conjunction with a number of modeled tumors heterogeneous in clonogen density, oxygenation, or proliferation. Models based on simple mathematical functions, published biological data, and biological image data were investigated. Target voxels were assigned to dose compartments using (i) simple rules based on the initial biological distribution, (ii) iterative methods designed to maximize the achievable TCP, or (iii) methods based on an ideal dose prescription. The relative performance of the simple rules was found to depend on the form of heterogeneity of the tumor, while the iterative and ideal dose methods performed comparably for all models investigated. In all cases the maximum achievable TCP was approached within the first few (typically two to five) compartments. Results suggest that irrespective of the pattern of heterogeneity, the optimal dose prescription can be well approximated using only a few dose levels but only if both the compartment boundaries and prescribed dose levels are well chosen.

A theoretical framework for prescribing radiotherapy dose distributions using patient-specific biological information.

We present a formalism for using functional imaging both to derive patient-specific radiobiological properties and consequently to prescribe optimal nonuniform radiotherapy dose distributions. The ability to quantitatively assess the response to an initial course of radiotherapy would allow the derivation of radiobiological parameters for individual patients. Both an iterative optimization and an analytical approach to this problem were investigated and illustrated by application to the linear-quadratic model of cell killing using simulated parametric data for a modeled tumor. Potential gains in local control were assessed by comparing uniform dose distributions with optimized dose distributions of equal integral dose. The effect on local prescribed dose of variations in effective radiosensitivity, tumor burden, and proliferation rate was investigated, with results suggesting that dose variations would be significant but clinically achievable. The sensitivity of derived parameters to image noise and the effect of varying the initial fractionation and imaging schedule were assessed. The analytical approach proved remarkably robust, with 10% image noise resulting in dose errors of approximately 1% for a clinically relevant set of parameters. Potential benefits were demonstrated by using this formalism to prescribe nonuniform dose distributions for model tumors using a range of literature-derived parameters. The redistribution of dose improved tumor control probability by factors between 1.03 and 4.27 for a range of model tumors.

A comparison of treatment planning techniques used in two randomised UK external beam radiotherapy trials for localised prostate cancer.

AIMS: To compare the radiotherapy planning techniques from two multicentre randomised external beam radiotherapy trials in the UK of conformal radiotherapy vs intensity-modulated radiotherapy (IMRT). MATERIALS AND METHODS: Sixteen sequential patients with histologically confirmed localised prostate cancer treated in the conventional or hypofractionated IMRT trial (CHHiP) were planned using both the CHHiP and Medical Research Council RT-01 planning protocols to 74 Gy in 37 daily fractions. The CHHiP plan used a single phase simple forward planned three-field IMRT plan for easy multicentre adoption. The RT-01 plan used two phases: three-field conformal radiotherapy plan to 64 Gy followed by a six-field boost of 10 Gy. After coverage of the planning target volumes according to the respective trial protocols, the dose to the rectum and bladder was assessed for the two planning techniques. RESULTS: There was acceptable planning target volume coverage by both the CHHiP and RT-01 plans. All CHHiP plans produced lower mean irradiated rectal volumes at all measured dose levels compared with the RT-01 plans, particularly for irradiated rectal volumes at 50 and 70 Gy (P<0.05). In the cases when a CHHiP plan failed to meet its own trial dose constraints, the volumes of irradiated rectum were less than if an RT-01 planning technique had been used. The CHHiP plans gave lower mean irradiated bladder volumes at both 50 and 60 Gy, but higher volumes at 74 Gy. These differences in irradiated bladder volumes were significant at the 60 and 74 Gy dose levels (P<0.05) in favour of the CHHiP and RT-01 plans, respectively. CONCLUSION: The forward planned CHHiP IMRT planning solution gives more favourable rectal sparing than the RT-01 plan. This is important to limit any potential increase in late rectal toxicity for prostate cancer patients treated with high-dose conventional or hypofractionated schedules.

Distortion-corrected T2 weighted MRI: a novel approach to prostate radiotherapy planning.

The purpose of this study was to evaluate distortion-corrected MRI as a radiotherapy planning tool for prostate cancer and the resultant implications for dose sparing of organs at risk. 11 men who were to be treated with radical conformal radiotherapy for localized prostate cancer had an MRI scan under radiotherapy planning conditions, which was corrected for geometric distortion. Radiotherapy plans were created for planning target volumes derived from the MRI- and CT-defined prostate. Dose volume histograms were produced for the rectum, bladder and penile bulb. The mean volume of the prostate as defined on CT and MRI was 41 cm3 and 36 cm3, respectively (p = 0.009). The predicted percentage of the rectum treated to dose levels of 45-65 Gy was significantly lower for plans delineating the prostate with MRI than for those with CT. The rectal-sparing effect was confined to the lowermost 4 cm of the rectum (anal canal). There were no differences between the predicted doses to bladder or penile bulb (as defined using MRI) between plans. In conclusion, prostate radiotherapy planning based on distortion-corrected MRI is feasible and results in a smaller target volume than does CT. This leads to a lower predicted proportion of the rectum, in particular the lower rectum (anal canal), treated to a given dose than with CT.