RadioBio data: A Moddicom Module to Predict Tumor Control

Probability and Normal Tissue Complication Probability in

Radiotherapy

Nicola Dinapoli

, Anna Rita Alitto

, Mauro Vallati

, Rosa Autorino

, Roberto Gatta

, Luca Boldrini

Andrea Damiani

, Giovanna Mantini

and Vincenzo Valentini

Department of Radiation Oncology, Universit

a Cattolica del Sacro Cuore, Rome, Italy

School of Computing and Engineering, University of Huddersﬁeld, Huddersﬁeld, U.K.

Keywords:

Dose Volume Histograms, Radiobiology, Radiotherapy, TCP, NTCP, Vdose, Dvolume.

Abstract:

In this work a system for analysing radiotherapy treatment planning dose-volume data is proposed. The work

starts from the deﬁnition of a framework inside a statistical scripting environment (R) used for creating a

software package. The analysis of dose-volume data in radiotherapy of malignant tumours is mandatory for

evaluating the prescribed treatment and for feedback analysis of outcome, both in the direction of tumour

control and in detection of parameters for estimating and predicting toxicity outcome. The statistical analysis

of large amount of clinical data can be slowed by the lack of practice in statistical tools needed, by clinicians,

to perform such kind of analysis. This is the reason that lead our working group in the creation of such a tool.

Finally an example of clinical application of our software is given for the analysis of the outcome of patients

undergoing to radiotherapy for prostate cancer.

1 INTRODUCTION

Radiotherapy is one of the most efﬁcient and most

used treatments in cancer therapy, together with the

chemotherapy and surgery. In the external beam Ra-

diotherapy (EBR) a radiation source is addressed to

send radiation beams (X-ray, gamma-rays or elec-

trons) on the patient, in speciﬁc locations, and the

leafs of a collimator are conﬁgured in order to repro-

duce in the space a desired 3-dimensional conforma-

tion of radiation doses. Such doses are planned in

order to be able to kill or severely damage the tumour

cells. The 3-dimensional shape of toxic dose is deliv-

ered inside the patient body; therefore, attention must

be payed in order to decrease as much as possible the

delivery of radiations in healthy tissues, so reducing

the likelihood of toxicity.

When planning a radiotherapy treatment, the radi-

ation oncologists have two critical aspects to consider:

(i) providing a total dose on the target that is adequate

for damaging and killing; and (ii) limiting the dose

received by normal tissue, in order to minimise iatro-

genic side effects. Usually, radiation oncologists have

to identify a reasonable trade-off between these two

aspects. Because of that, physicians exploit complex

mathematical models – called radiobiological mod-

els – for estimating the Tumour Control Probability

(TCP) and Normal Tissue Complication Probability

(NTCP). Given such estimations, they can identify

the plan that provides the largest total dose on target,

while limiting the likelihood of normal tissue side ef-

fects.

Despite the fact that numerous models have been

proposed, the identiﬁcation and exploitation of radio-

biological models is still considered an open and criti-

cal topic. Such models are currently in the aims of the

most important international associations in the ﬁeld:

for example, the American Society for Radiation

Oncology (ASTRO) in association with the Ameri-

can Association of Physicists in Medicine (AAPM)

supported the creation of Quantitative Analyses of

Normal Tissues Effects in the Clinic (QUANTEC).

QUANTEC project exploits the large amount of data

that modern technologies – like Linear Accelerator

(LINAC) and Treatment Planning Systems (TPSs) –

are able to export, in order to provide more accurate

TCP and NTCP speciﬁc for the different involved hu-

man anatomical regions (Bentzen et al., 2010; Marks

et al., 2010b).

One of the most common source of information

exploited by radiobiological models is represented by

Dinapoli, N., Alitto, A., Vallati, M., Autorino, R., Gatta, R., Boldrini, L., Damiani, A., Mantini, G. and Valentini, V.

RadioBio data: A Moddicom Module to Predict Tumor Control Probability and Normal Tissue Complication Probability in Radiotherapy.

DOI: 10.5220/0005693502770281

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 5: HEALTHINF, pages 277-281

ISBN: 978-989-758-170-0

277

the Dose Volume Histogram (DVH): an histogram

showing on the x-axes the dose and on the y-axes

the volume of the ROI receiving such dose. Each

DVH is a “quick 2D recap”, of the dose absorbed for

each target or normal tissue volume, also called Re-

gion Of Interest (ROI). In literature, the analysis of

DVHs is related to the actual clinical outcome, for ef-

fectively identifying some dose constraints under the

form of clear rules, such as: the minimum dose to the

Planning Target Volume is 95% of the plan reference

point dose; in order to decrease risk of radiation pneu-

monia under 20% it is prudent to limit Mean Lung

Dose within 20 Gy (Marks et al., 2010a) or “to avoid

rectal bleeding or late rectal RTOG grade >= 2 less

than 50% of rectal volume should receive 50 Gy or

less than 35% should receive 60 Gy”(Michalski et al.,

2010).

The classical empirical approach to deﬁne such

constrains has been recently overcome by adaptive al-

gorithms, which are able to adaptively ﬁnd, with the

evidence of the daily clinical practice, also new form

of useful constraints. An important contribute come

from (Naqa et al., 2006) which clearly highlights the

importance of predictors built basing on DVH and

a multi-variate modelling process. Nevertheless all

those experiences are done locally, in different centres

with different tools, and there is no common tools to

build and tune new TCP/NTCP models by a speciﬁc

dataset.

Despite collecting and analysing dose information

is considered very important in order to build and im-

prove maths models, there is a poor availability of eas-

ily to access tools for such task. VODCA is the oldest

project and is able to collect DICOM-RT studies and

to analyse them by using R, a well-known statistical

engine. It should be noted that VODCA is not free and

does not provide speciﬁc tools, libraries and function

for TCP/NTCP models.

In order to provide a common tool for building

and ﬁnely tuning TCP/NTCP models, in this paper

we propose a software library, integrated in a more

complete project called moddicom(Dinapoli et al.,

2015b), that is able to effectively support the process

of generating and improving TCP/NTCP models for

the speciﬁc data of a radiotherapy centre. To foster the

exploitation of the tool, and therefore allow a better

use of the knowledge stored in the existing datasets,

the proposed software is freely available as an R pack-

age at https://github.com/kbolab/moddicom.git. In

this paper we provided an overview of the developed

software library, and we show the results of an exper-

imental analysis focused on evaluating the usefulness

of the proposed system.

2 METHODS

The overall structure of the moddicom framework is

shown in Figure 1. Currently, the modules grouped as

R environment are available and ready to be used. The

R environment includes:

• geoLet. This requires the DCMTK Ofﬁce li-

braries (Eichelberg et al., 2004) and allows to load

DICOM Images, DICOM RTStruct, DICOM RT-

Dose, DICOM RTPlan studies in memory.

• Image Feature Data Analysis. This module al-

low to extract image features for Radiomics (Di-

napoli et al., 2015a) analysis from the previously

loaded DICOM studies.

• RadioBio data models. This module cal-

culates DVHs (starting from the DICOMRT

Struct/Dose/Plan previously loaded by geoLet)

and can tune the parameters of six different radio-

biological maths models for TCP/NTCP mod-

elling.

• Visualisation Module. This module provides

tools to visualise objects and images in 2D/3D in

order to give a practical presentation of the stored

data.

Part of the moddicom framework is still under de-

velopment. The complete moddicom framework will

be able to acquire information directly from modali-

ties and PACS, via DICOM protocol, in order to popu-

late its internal database, which will be subsequently

able to automatically exploit the stored information.

At the moment, the loading is done by searching a

ﬁle system and for ﬁnding DICOM information.

The geoLet module can search the ﬁle system,

from a given path, and load the data stored in

DICOM RT-PLAN/RT-STRUCT and RT-DICOM-

Distribution Doses. The corresponding information

are then used by the RadioBio data module for gen-

erating and ﬁtting a number of mathematical models.

The ﬁtting is also able to analyse the hidden clinical

effects of dose constraint not yet identiﬁed as signif-

icant in literature. In the following, the implemented

models are listed and brieﬂy described. The interested

reader is referred to the provided references for details

about the mathematical functions and models.

• Lyman. It is a reformulation of probit model, that

uses different parameters than mean and standard

deviation (Burman et al., 1991).

• Goitein. This model is similar to the Lyman

model, but it is function of logarithm of the Dose

(Shipley et al., 1979; Bentzen and Tucker, 1997).

• Niemierko. It is the translation of a loglogit gen-

eralised linear model as function of T D

and γ

(Gay and Niemierko, 2007).

HEALTHINF 2016 - 9th International Conference on Health Informatics

278

Figure 1: The overall structure of the moddicom architecture. The core is represented in the box titles R environment and is

currently available. The structure includes also the PACS/TPS interface and the web based GUI.

• Munro. This model is an empirical dose/response

curve that ﬁts experimental data. The model is

based on the assumption that this curve is equiv-

alent to a Poisson model (Munro and Gilbert,

1961).

• Okunieff. This model is the equivalent of the lo-

gistic generalised linear model, where the covari-

ates and their coefﬁcients have been reported as

function of T D

and γ

(Okunieff et al., 1995).

• Warkentin. This model is the equivalent of the

Warkentin Poisson model where the covariates

and their coefﬁcients have been reported as func-

tion of T D

and γ

(Warkentin et al., 2004).

• Bentzen. Bentzen described dose/response

curves both for tumour and normal tissue re-

sponse. In our implementation, log Poisson model

starts from equation used in Warkentin equation,

but the subsequent evaluation of the relation-

ship of response with the log Dose (Bentzen and

Tucker, 1997).

The previously cited models are considered in

order to automatically ﬁnd the most signiﬁcant

Dose/Volume constraint able to ﬁt the model with the

clinical outcomes. A similar approach has been pro-

posed in the past (Naqa et al., 2006), but it is worthy

noting that moddicom represents the ﬁrst shared and

freely available software tool able to do that.

The RadioBio data module also provide a simula-

tion environment where DVH can be artiﬁcially cre-

ated (Van den Heuvel, 2006) and associated to clinical

outcome according with a known statistic. This repre-

sents an useful sandbox to test the implemented math-

ematical models and see their details in action. As

previously cited, RadioBio data module is also linked

with the geoLet module for being able to work with

real clinical data directly taken from DICOM sources.

For providing a complete environment to the re-

searches, a range of tools –including calibration plots

and ﬁtting curve– have been implemented and can be

exploited within the moddicom framework.

3 EXPERIMENTAL RESULTS

In order to test effectively the applicability and the

capability of the proposed RadioBio data moddicom

module, a clinical investigation was performed on

a real dataset of 123 patients treated with Radia-

tion Therapy for prostate cancer: this cohort was

taken from a study performed between September

2010 and May 2014 at the Radiation Oncology Cen-

tre of Gemelli Hospital. Acute and late rectal (GI)

and bladder (GU) toxicity data were collected. Re-

ferring physician clinically examined patients a me-

dian of 5 times during RT treatment, unless oth-

erwise requested by patient himself. RTOG score

(Cox et al., 1995) and CTC-AE v.4.03 (of Health

and Human Services, 2010) were used to stratify

grade of toxicity. Patients’ DVH were collected to

evaluate bladder and rectum dose-volume distribu-

tions. Dose from treatment planning and simulated

delivery was evaluated using the Equivalent Uniform

Dose (EUD) (Niemierko, 1997). For normal tissues,

the EUD represents the uniform dose which leads to

the same probability of injury as the examined inho-

mogeneous dose distribution. The EUD was calcu-

lated from the corresponding dose-volume distribu-

tions (histograms).

To calculate the EUD-based normal tissue com-

plication probability (NTCP), we used parametrisa-

tion of the dose-response characteristics using Lyman

RadioBio data: A Moddicom Module to Predict Tumor Control Probability and Normal Tissue Complication Probability in Radiotherapy

279

Figure 2: Correlation between rectal G1 late toxicity and a

Vdose=59.5 Gy.

Figure 3: Correlation between rectal G1 late toxicity and a

Dvolume=16.44cc

NTCP formula (Burman et al., 1991).

In such comparison DVHs have been clustered ac-

cording with the induced toxicity (Figure 4). The

dose-volume toxicity parametrisation has been con-

ducted by identifying respectively the volume of a

structure receiving a given level of dose (as known as

Vdose) and the dose received by a given volume of the

structure (as known as Dvolume). Such dose-volume

parametrisation is extensively in radiotherapy for cor-

relating outcome and dose distribution data (Marks

et al., 2010b). We observed a correlation between

development of late G1 or superior GI toxicity and

a Vdose equal to 59,5 Gy (see Figure 2) and a Dvol-

ume equal to 16,44 cc (see Figure 3). In particular we

found a signiﬁcant correlation with D16cc by Mann-

Whitney Test (p = 0.019).

Finally, to elaborate a predictive model for G1 GI

toxicity, we build up a DVH-reduction model by Ly-

man Model Maximum Likelihood Estimation, based

on estimated NTCP under uniform irradiation (EUD)

of the rectum. We found that our patients were dis-

tributed on or close to an S slope showing that 50%

probability of acute toxicity is present with a mean

delivered uniform dose of 66,5 Gy (see FIgure 4).

The results of the analysis, performed using the

Figure 4: The DVH-reduction model based on estimated

complication probability (NTCP) under uniform irradiation

(EUD) of the rectum.

Figure 5: The proposed graphical user interface.

proposed moddicom module, conﬁrmed the evidence

known in literature and also suggested a new possi-

ble signiﬁcant dose constraint. While the former re-

sult clearly indicates the validity of the approach, the

latter provides some insights into the usefulness of

the RadioBio data module. However, due to the lim-

ited number of recruited patients, further investigation

should be done to conﬁrm the identiﬁed constraint.

Remarkably, the analysis was performed by a

physician –although expert user of R– rather than a

technician. Therefore, the use of moddicom can pro-

vide an effective support to physician in their every

day research or clinical activities.

4 CONCLUSIONS

In this paper we presented the RadioBio data module

of moddicom, which is a free and open source library

born to support the adaptive TCP/NTCP modelling.

We tested RadioBio data with 123 patients treated

with Radiation therapy for prostate cancers and we

obtained results in accord with literature. The per-

formed analysis focused on investigating the usability

and usefulness of the proposed module, and it showed

HEALTHINF 2016 - 9th International Conference on Health Informatics

280

that RadioBio data can be an useful tool to mine ef-

fective dose constraints hidden into the shape of Dose

Volume Histograms. Moreover, the considered case-

study can be extended by considering a larger sample

of patients, in order to provide stronger evidences and

better optimise planning procedures in treatment val-

idation and predict possible toxicity. The future chal-

lenge will be the personalisation of treatments and

complications rates reduction.

Future work include the development of a user-

friendly graphical user interface, as proposed in Fig-

ure 5, an experimental analysis considering a larger

number of patients, and the involvement of a larger

number of medical experts.

REFERENCES

Bentzen, S. and Tucker, S. (1997). Quantifying the position

and steepness of radiation dose-response curves. Int J

Radiat Biol., 71(5):531–42.

Bentzen, S. M., Constine, L. S., Deasy, J. O., Eisbruch, A.,

Jackson, A., Marks, L. B., Haken, R. K. T., and Yorke,

E. D. (2010). Quantitative analyses of normal tissue

effects in the clinic (quantec): An introduction to the

scientiﬁc issues. Radiation Oncology, 76(3):S3–S9.

Burman, C., Kutcher, G., Emami, B., and Goitein, M.

(1991). Fitting of normal tissue tolerance data to

an analytic function. Int J Radiat Oncol Biol Phys,

21(1):123–35.

Cox, J., Stetz, J., and PajaK, T. (1995). Toxicity criteria

of the radiation therapy oncology group (rtog) and the

european organization for research and treatment of

cancer (eortc). Int. J. Radiation Oncology Biol. Phys,

31(5):1341–1346.

Dinapoli, N., Alitto, A., Vallati, M., Gatta, R., Autorino, R.,

Boldrini, L., Damiani, A., and Valentini, V. (2015a).

Moddicom: A complete and easily accessible library

for prognostic evaluations relying on image features.

In IEEE Engineering in Medicine and Biology Society,

37th Annual International Conference.

Dinapoli, N., Alitto, A. R., Vallati, M., Gatta, R., Au-

torino, R., Boldrini, L., Damiani, A., and Valentini,

V. (2015b). Moddicom: a complete and easily ac-

cessible library for prognostic evaluations relying on

image features. In Proceedings of the 37th Annual

International Conference of the IEEE Engineering in

Medicine and Biology Society (EMBS). IEEE.

Eichelberg, M., Riesmeier, J., Wilkens, T., Hewett, A.,

Barth, A., and Jensch, P. (2004). Ten years of medical

imaging standardization and prototypical implemen-

tation: the dicom standard and the ofﬁs dicom toolkit

(dcmtk). In Medical Imaging 2004, pages 57–68.

Gay, H. and Niemierko, A. (2007). A free program for cal-

culating eud-based ntcp and tcp in external beam ra-

diotherapy. Phys Med., 23(3-4):115–25.

Marks, L., Bentzen, S., Deasy, J., Kong, F.-M., Bradley,

J., Vogelius, I., Naqa, I., Hubbs, J., Lebesque, J., Tim-

merman, R., Martel, M., and Jackson, A. (2010a). Ra-

diation dose-volume effects in the lung. Int. J. Radia-

tion Oncology Biol. Phys, 76(3):S70–S76.

Marks, L. B., Yorke, E. D., Jackson, A., Haken, R. K. T.,

Constine, L. S., Eisbruch, A., Bentzen, S. M., Nam, J.,

and Deasy, J. O. (2010b). Use of normal tissue com-

plication probability models in the clinic. Radiation

Oncology, 76(3):S3–S9.

Michalski, J., Gay, H., Jackson, A., Tucker, S., and Deasy,

J. (2010). Radiation dose-volume effects in radiation-

induced rectal injury. Int. J. Radiation Oncology Biol.

Phys, 76(3):S123–S129.

Munro, T. R. and Gilbert, C. W. (1961). The relation be-

tween tumour lethal doses and the radiosensitivity of

tumour cells. Br J Radiol., 34:246–51.

Naqa, I. E., Bradley, J., Blanco, A., Lindsay, P., Vicic, M.,

Hope, A., and Deasy, J. (2006). Multi-variable model-

ing of radiotherapy outcomes including dose-volume

and clinical factors. Int. J. Radiation Oncology Biol.

Phys, 64(4):1275–1286.

Niemierko, A. (1997). Reporting and analyzing dose distri-

butions: a concept of equivalent uniform dose. Medi-

cal physics, 24(1):103–10.

of Health, U. D. and Human Services, National Institutes

of Health, N. C. I. v. (June 14 2010). Common termi-

nology criteria for adverse events (ctcae).

Okunieff, P., Morgan, D., Niemierko, A., and Suit, H.

(1995). Radiation dose-response of human tumors.

Int J Radiat Oncol Biol Phys, 32(4):1227–37.

Shipley, W. U., Tepper, J. E., Prout, G. R., Verhey, L., Men-

diondo, O., Goitein, M., Koehler, A., and Suit, H.

(1979). Proton radiation as boost therapy for local-

ized prostatic carcinoma. JAMA, 241(18):1912–5.

Van den Heuvel, F. (2006). Decomposition analysis of dif-

ferential dose volume histograms. Medical Physics,

33(2):297.

Warkentin, B., Stavrev, P., Stavreva, N., Field, C., and Fal-

lone, B. (2004). A tcp-ntcp estimation module using

dvhs and known radiobiological models and parame-

ter sets. J Appl Clin Med Phys, 5(1):50–63.

RadioBio data: A Moddicom Module to Predict Tumor Control Probability and Normal Tissue Complication Probability in Radiotherapy

281