• Nenhum resultado encontrado

Identification and application of image biomarkers for the prediction of radiotherapy treatment response in head and neck cancer patients

N/A
N/A
Protected

Academic year: 2021

Share "Identification and application of image biomarkers for the prediction of radiotherapy treatment response in head and neck cancer patients"

Copied!
113
0
0

Texto

(1)

FACULDADE DE CIÊNCIAS

DEPARTAMENTO DE FÍSICA

Identification and Application of Image Biomarkers for

the Prediction of Radiotherapy Treatment Response in

Head and Neck Cancer Patients

 

Cássia Oraboni Ribeiro

Orientadores

Dra. Marianna Sijtsema, Física médica, Department of Radiation Oncology of the

University Medical Center Groningen, Groningen, The Netherlands

Prof. Dr. Luis Peralta, Professor associado, Departamento de Física da Faculdade de

Ciências da Universidade de Lisboa, Lisboa, Portugal

 

Mestrado Integrado em Engenharia Biomédica e Biofísica

Perfil Radiações em Diagnóstico e Terapia

 

DISSERTAÇÃO

-2015-

(2)

     

      ii    

(3)

      iii  

Acknowledgments

 

  First  of  all,  I  would  like  to  express  my  sincere  gratitude  to  all  the  people,  both  from   the   University   in   Portugal   and   the   University   Medical   Center   Groningen   (UMCG),   that   allow  for  the  possibility  of  this  internship  and  the  help  provided  in  the  arrangements.  This   was  undoubtedly  a  unique  life  experiencing,  which  made  me  a  different  and  better  person,   both  personally  and  professionally.  

  The  chair  of  the  department  of  Radiation  Oncology  of  the  UMCG,  Prof.  Dr.  Johannes   Langendijk,  was  the  first  person  I  made  contact  with  for  the  possibility  of  this  internship.  I   am   extremely   thankful   for   all   his   availability   and   the   excellent   conditions   he   ensured   during  my  stay.    

Dr.   Marianna   Sijtsema   and   Dr.   Roel   Steenbakkers   gave   me   the   opportunity   of   working   in   the   department   and   without   them   this   internship   would   not   have   been   possible.   Both   of   them   were   always   a   great   support   and   their   professionalism   and   motivation  in  supervising  me  were  always  present.  I  am  also  sincerely  grateful  to  Sanne   van   Dijk   for   all   her   guidance   and   help.   Without   her   presence,   this   project   would   not   be   possible   at   all.   I   also   would   like   to   thank   all   the   staff   from   the   department,   they   were   always  kind  and  available.  I  am  happy  to  say  that  all  of  them  turned  out  to  be  very  special   people  in  my  life  and  I  am  grateful  for  their  friendship.  Prof.  Luis  Peralta,  as  the  internal   supervisor,  was  always  present  and  his  rapid  response  and  concern  was  outstanding.  I  am   also   thankful   to   Marianna,   Roel,   Sanne   and   Prof.   Luis   for   all   the   patience,   support   and   guidance  during  the  execution  of  this  thesis.    

Inês   Lourenço   and   Patricia   Queiroz   turned   this   nine   months   in   The   Netherlands   absolutely  amazing  and  succeeded  in  making  this  experience  one  of  the  most  important  of   my   life.   I   am   sincerely   grateful   for   their   support   and   true   friendship.   Also   my   boyfriend,   Nelson   Afonso,   was   fundamental   for   all   my   achievements   and   without   him   everything   would  be  more  difficult.  

A  very  special  affection  goes  to  my  role  models,  my  parents,  without  whom  I  would   not  have  the  possibility  to  undergo  for  all  the  amazing  experience  of  working  and  living   abroad.  I  am  also  very  grateful  for  all  their  support  during  this  period  from  Portugal.    

 

(4)

     

(5)

      v  

Resumo

  A  radioterapia  é  uma  modalidade  de  tratamento  de  cancro  amplamente  utilizada   em   todo   o   mundo.   Este   tipo   de   terapia   tem   como   principal   objectivo   focalizar   uma   alta   dose   de   radiação   para   o   tumor   e   ao   mesmo   tempo   minimizar   a   exposição   dos   tecidos   saudáveis  presentes  à  volta  do  mesmo.      

A   concretização   deste   tipo   de   tratamento   com   alguma   precisão   e   eficiência   já   se   torna  possível  graças  às  mais  modernas  técnicas,  tais  como  a  Radioterapia  Conformacional   em   3D   (3D-­‐CRT),   Radioterapia   de   Feixe   de   Intensidade   Modulada   (IMRT)   e   Terapia   Arc   Modulado   Volumétrico   (VMAT).   O   surgimento   destas   técnicas   foi   possível   devido   à   modernização   de   modalidades   de   imagem   usadas   em   oncologia.     Mais   concretamente,   deve-­‐se   ao   surgimento   da   Tomografia   Computadorizada,   usualmente   conhecida   por   CT   (do  inglês  Computed  Tomography),  que  é  o  tipo  de  imagem  principal  e  mais  utilizada  em   radioterapia.    

Contudo,   apesar   de   todos   estes   avanços,   atualmente   ainda   não   é   possível   prever   com   precisão   o   resultado   de   um   determinado   tratamento   de   radioterapia   para   um   paciente  específico.    Seria  bastante  útil  conseguir  obter  esta  informação  mesmo  antes  de   se   sujeitar   o   paciente   à   radiação,   por   forma   a   adaptar   o   tratamento   de   acordo   com   as   necessidades   e   características   da   doença.   Esta   prática   designa-­‐se   por   medicina   personalizada   e   permite   resultados   muito   mais   promissores   para   uma   determinada   terapia.  

A   CT   tem   a   vantagem   de   ser   um   método   não   invasivo.   É   possível   através   das   imagens   a   3   dimensões   produzidas   conseguir-­‐se   distinguir   (visualmente)   diferentes   características  de  um  tumor,  relativamente  à  sua  intensidade,  forma,  tamanho  ou  textura.   Neste  contexto,  surge  um  campo  emergente  em  processamento  de  imagem  designado  por  

radiomics.   Este   método   permite   caracterizar   uma   imagem   quantitativamente,   o   que  

possibilita   a   definição   muito   mais   precisa   do   fenótipo   de   um   determinado   tumor.   Radiomics   trata   da   extração   das   designadas,   ao   longo   desta   tese,   características   quantitativas  de  uma  imagem.  

  Assim,  neste  projeto  de  estágio  foi  estudada  a  possibilidade  de  se  conseguir  prever   a   eficiência   de   um   tratamento   de   radioterapia   através   da   aplicação   de   radiomics   em   imagens  de  CT.  A  população  em  estudo  foram  os  doentes  com  carcinoma  espinocelular  de   cabeça  e  pescoço  (HNSCC).  A  eficiência  do  tratamento  foi  estudada  através  da  análise  de   sobrevivência   de   310   pacientes   portadores   desta   doença   e   que   fizeram   tratamento   com  

(6)

     

      vi    

radioterapia   no   University   Medical   Center   Groningen   (UMCG).   Este   centro   hospitalar,   localizado  na  cidade  de  Groningen  (no  norte  da  Holanda),  foi  o  local  do  estágio  efectuado   por  9  meses  para  a  realização  deste  projeto.  

Utilizou-­‐se  o  Matlab  para  implementar  125  algoritmos  diferentes  de  características   de  imagem.  Posteriormente,  os  valores  destas  características  de  imagem  foram  extraídos   das   imagens   de   CT   dos   tumores   dos   310   pacientes   analisadas.   O   software   de   estatística   designado  por  R  foi  o  programa  maioritariamente  utilizado  na  parte  da  análise  dos  valores   obtidos  das  características  de  imagem  para  os  pacientes  deste  estudo.  

As   características   de   imagem   implementadas   foram   divididas   em   3   grupos   diferentes.   O   primeiro   grupo   consiste   no   cálculo   de   métricas   a   partir   do   histograma   de   intensidades  de  todos  os  voxels  da  imagem  de  CT  que  contém  o  tumor.  O  segundo  grupo     traduz   a   forma   e   tamanho   do   tumor   e   por   fim,   o   último   e   terceiro   grupo,   quantifica   a   heterogeneidade   do   mesmo.   Apenas   as   características   de   imagem   implementadas   que   mostrem   uma   diferença   significativa   (entre   valores   baixos   e   altos   dessa   mesma   característica)   na   sobrevivência   após   o   início   do   tratamento   serão   investigadas   em   mais   pormenor  e  designados  ao  longo  desta  tese  por  biomarcadores  de  imagem.    

O   objectivo   deste   projeto   é   encontrar   estes   biomarcadores   de   imagem   e   determinar   o   seu   desempenho   na   previsão   do   resultado   de   um   tratamento   de   radioterapia.    

Em   publicações   anteriores,   já   se   provou   a   relação   de   certas   características   quantitativas  de  uma  imagem,  que  traduzem  especificações  do  tumor,  com  a  sobrevivência   [1].  Por  isso,  numa  primeira  fase  da  análise  estatística  deste  projeto,  alguns  dos  resultados   já   obtidos   por   trabalhos   anteriores   foram   validados   com   os   pacientes   disponíveis   para   este  estudo.    

A  mais-­‐valia  que  esta  tese  traz  para  o  campo  de  radiomics  é  a  análise  da  influência   dos   valores   das   características   de   imagem   na,   para   além   da   sobrevivência   no   geral,   recorrência  do  tumor  e  no  aparecimento  de  metástases.  Novas  e  diferentes  características   de  imagem  foram  também  implementadas.    

  Após   o   início   da   análise   dos   dados   ao   longo   do   estágio,   foram   encontrados   55   biomarcadores   de   imagem   (44%   das   125   características   de   imagem   implementadas),   o   que  mostra  o  potencial  do  conceito  de  radiomics  na  prática  clínica  de  radioterapia.    

Para   além   disso,   através   de   um   método   eficiente   de   seleção   de   variáveis,   pretendeu-­‐se   escolher   as   3   melhores   características   de   imagem,   de   forma   a   combiná-­‐las   num  modelo  capaz  de  prever  a  sobrevivência  de  pacientes  com  carcinoma  espinocelular   de   cabeça   e   pescoço   tratados   com   radioterapia.   O   método   de   seleção   garantiu   que   as  

(7)

      vii   características   de   imagem   selecionadas   tivessem   valores   estáveis,   não   fossem   correlacionadas   entre   si   e   apresentassem   (individualmente   e   em   combinação)   os   mais   altos  valores  de  desempenho  de  prognóstico  possíveis.  

Seguindo   este   método,   os   3   biomarcadores   de   imagem   selecionados   designam-­‐se   por:   Uniformity,   Surface   density   e   Size   zone   non-­‐uniformity,   correspondestes   ao   primeiro,   segundo   e   terceiro   grupo,   respectivamente.   Estas   3   características   em   combinação   designam-­‐se  por  assinatura  de  biomarcadores  de  imagem.  O  desempenho  de  um  modelo   contendo  estes  3  biomarcadores  de  imagem  foi  muito  promissor  e  após  validação  interna   obteve-­‐se   um   valor   de   aproximadamente   0.65   (numa   escala   de   0.5   até   1),   que   é   considerada  uma  performance  de  prognóstico  muito  promissora.  

De   forma   a   analisar   o   valor   suplementar   de   características   quantitativas   de   imagem  na  prática  de  radioterapia,  foram  comparados  o  desempenho  de  prognóstico  da   assinatura   de   biomarcadores   de   imagem   com   inúmeros   parâmetros   específicos   de   pacientes  já  comummente  utilizados  na  clínica.  A  assinatura  de  biomarcadores  de  imagem   obteve  um  desempenho  superior  a  todos  os  parâmetros  excepto  o  designado  WHO,  que  se   trata  de  uma  pontuação  dada  pelo  médico  antes  do  tratamento  que  incide  sobre  o  estado   físico   do   paciente   (se   é   totalmente   independente   e   ativo).   É   natural   que   este   parâmetro   consiga   prever   minuciosamente   o   resultado   do   tratamento   visto   que   a   eficiência   do   mesmo  depende  bastante  das  condições  físicas  prévias  do  paciente.  

Para  se  conseguir  um  melhor  poder  de  prognóstico,   decidiu-­‐se  então  combinar  a   assinatura   de   biomarcadores   de   imagem   com   a   pontuação   WHO   num   modelo   designado   por   modelo   final.   Com   o   modelo   final,   obteve-­‐se   uma   performance   de   aproximadamente   0.73,  que  indica  uma  capacidade  excelente  de  predição.  

Finalmente,  após  a  obtenção  de  uma  assinatura  de  biomarcadores  de  imagem  e  de   um   modelo   final,   desenvolveu-­‐se   sistemas   de   estratificação   de   risco   pré-­‐tratamento,  que   acaba   por   ser   mais   uma   das   grandes   novidades   que   este   projeto   traz   para   o   campo   de  

radiomics  no  geral.  Estes  sistemas  baseiam-­‐se  na  assinatura  de  biomarcadores  de  imagem  

e  no  modelo  final  para  diferenciar  (previamente  ao  tratamento)  os  pacientes  considerados   com   baixo   risco   e   com   alto   risco   de   morte,   recorrência   de   tumor   ou   metástases.   As   diferenças  obtidas  para  os  dois  grupos  de  risco  foram  bastante  significativas  em  qualquer   um   dos   casos   (morte,   recorrência   ou   metástases).   Neste   projeto,   estes   sistemas   de   risco   foram   desenvolvidos   com   a   informação   dos   pacientes   disponíveis,   mas   a   ideia   é   serem   aplicados   a   um   conjunto   externo   de   doentes   pertencentes   à   mesma   população.   Isto   traz   inúmeros   benefícios   clínicos,   pois   torna-­‐se   possível   o   desenvolvimento   da   medicina  

(8)

     

      viii    

personalizada,  e  consequentemente,  o  aumento  das  taxas  de  sucesso  de  radioterapia  nos   pacientes  com  carcinoma  espinocelular  de  cabeça  e  pescoço.  

Os  resultados  obtidos  ao  longo  do  estágio  foram  bastante  promissores,  indicando  o   poder   de   prognóstico   das   características   de   imagem   implementadas.   Contudo,   torna-­‐se   necessário  validar  tanto  os  modelos  desenvolvidos  como  os  sistemas  de  estratificação  de   risco  pré-­‐tratamento  num  conjunto  externo  de  pacientes  com  carcinoma  espinocelular  de   cabeça   e   pescoço   por   forma   a   se   obter   um   desempenho   menos   optimista,   mas   possivelmente  mais  realista  e  mais  representativo  da  população  no  geral.  Este  será  talvez   o  próximo  passo  para  o  complexo  processo  da  introdução  de  biomarcadores  de  imagem  na   prática  clínica  de  radioterapia.  

 

Palavras-­‐chave:  radioterapia,  radiomics,  CT,  tumor,  biomarcadores  de  imagem.      

(9)

      ix    

Abstract

Each   year,   about   300   patients   with   head   and   neck   squamous   cell   carcinoma   (HNSCC)   are   treated   with   radiation   therapy   (either   primary   or   post-­‐operative)   at   the   department   of   Radiation   Oncology   of   the   University   Medical   Center   Groningen   (UMCG),   Groningen,  The  Netherlands.  To  our  knowledge,  this  hospital  has  the  largest  database  of   information  on  toxicity  in  the  world  for  the  head  and  neck  cancer  (HNC)  patients  treated   with  radiotherapy.  

In   literature   it   was   shown   that   specific   image   characteristics   (440   in   total)   of   a   tumour   are   correlated   to   its   patient   overall   survival   (OS)   [1].   Radiomics   is   the   field   that   allows   large   amounts   of   image   characteristics   (features)   to   be   extracted   automatically   from  radiographic  images.  The  features  that  demonstrate  a  correlation  to  survival  can  be   also  called  image  biomarkers.    

The  main  goal  of  this  work  is  to  investigate  in  more  detail  the  prognostic  value  of   image  features  in  the  planning  CT-­‐scans  of  HNC  patients  (310  in  total)  for  their  response   to  radiotherapy  (RT).  In  addition,  the  relation  between  the  image  features  and  recurrence-­‐ free  survival  (RFS)  will  be  assessed.    

Overall,   the   obtained   results   showed   the   promising   power   of   radiomics.   Particularly,   55   image   features   demonstrated   significant   differences   for   OS.   Moreover,   a   prognostic  performance  of  approximately  0.65  (scored  from  0.5  to  1)  was  obtained  for  the   selected   image   features   in   the   prediction   of   RT   outcome   for   a   HNC   patient   with   similar   characteristics   as   the   dataset   used   for   this   project.   Also   the   better   performance   of   the   image   biomarkers   in   relation   to   other   commonly   used   patient-­‐specific   parameters   is   shown.  Additionally,  the  ability  of  image  features  in  distinguishing  between  high  and  low   risk   patients   beforehand   is   demonstrated,   which   can   contribute   to   a   more   tailored   treatment,  and  consequently,  higher  success  rates.  

Therefore,  different  but  very  promising  results  were  obtained  when  compared  to   previous  work.  The  methodology  followed  for  this  project  revealed  the  promising  power   of  the  image  biomarkers  to  accurately  predict  the  RT  treatment  response  in  HNC  patients.   Therefore,   the   next   step   will   consist   in   validating   the   developed   models   in   external   datasets  of  patients  for  a  possible  clinical  use  in  the  future.    

 

Key  words:  radiotherapy,  radiomics,  CT,  tumour,  image  biomarkers.  

(10)

     

      x    

(11)

      xi    

Acronyms

2D:  Two-­‐Dimensional  

3D-­‐CRT:  3D  Conformal  Radiation  Therapy

 

3D:  Three-­‐Dimensional  

ANOVA:  ANalysis  Of  VAriance   AUC:  Area  Under  the  Curve   c-­‐index:  concordance  index  

CERR:  Computational  Environment  for  Radiotherapy  Research   CT:  Computed  Tomography  

CTV:  Clinical  Target  Volume   DVH:  Dose-­‐Volume  Histogram  

FDG-­‐PET:  18F-­‐FluorDesoxyGlucose  Positron  Emission  Tomography     FDG:  18F-­‐FluorDesoxyGlucose    

FDR:  False  Discovery  Rate    

GLCM:  Grey-­‐Level  Co-­‐occurrence  Matrix     GLRLM:  Grey-­‐Level  Run-­‐Length  Matrix   GLSZM:  Grey-­‐Level  Size-­‐Zone  Matrix   GTV-­‐LN:  GTV  of  the  Lymph  Nodes   GTV-­‐PT:  GTV  of  the  Primary  Tumour   GTV:  Gross  Tumour  Volume  

HNC:  Head  and  Neck  Cancer  

HNSCC:  Head  and  Neck  Squamous  Cell  Carcinoma   HPV:  Human  Papillomavirus      

HU:  Hounsfield  Units    

ICC:  Intraclass  Correlation  Coefficient  

IMRT:  Intensity-­‐Modulated  Radiation  Therapy   LINAC:  LINear  ACcelerator  

MLC:  MultiLeaf  Collimator  

MRI:  Magnetic  Resonance  Imaging  

NGTDM:  Neighbourhood  Grey-­‐Tone  Difference  Matrix   NSCLC:  Non-­‐Small  Cell  Lung  Carcinoma  

NTCP:  Probability  of  Normal  Tissue  Complications   OAR:  Organs  At  Risk  

(12)

     

      xii    

OS:  Overall  Survival  

PEG:  Percutaneous  Endoscopic  Gastrostomy     PET:  Positron  Emission  Tomography  

PH:  Proportional  Hazards     PTV:  Planning  Target  Volume   RFS:  Recurrence-­‐Free  Survival    

ROC:  Receiver  Operating  Characteristic     RT:  Radiation  Therapy  

SCC:  Squamous  Cell  Carcinoma   TCP:  Tumour  Control  Probability  

UICC:  Union  for  International  Cancer  Control     UMCG:  University  Medical  Center  Groningen   VMAT:  Volumetric  Modulated  Arc  Therapy   VOI:  Volume  Of  Interest  

(13)

     xiii    

List of Figures

Figure  2.1  -­‐  The  anatomical  subsites  of  the  upper  aerodigestive  tract  [70].  ...  4  

Figure  2.2  -­‐  Endoscopic  examination  of  a  SCC  of  the  larynx  [71].  ...  5  

Figure  2.3  -­‐  Stage  grouping  for  hypopharyngeal  cancer  [22].  ...  6  

Figure  2.4  -­‐  An  MLC  perpendicular  to  the  eye  view  of  the  beam  [11].  ...  9  

Figure  2.5  -­‐  Dose  distribution  in  a  typical  head  and  neck  IMRT  plan  [11].  ...  11  

Figure  2.6  -­‐  TCP  and  NTCP  curves  [72].  ...  12  

Figure   2.7   –   DVH   for   a   HNC   case   for   3   different   delineated   structures   in   a   radiotherapy   treatment   plan.   The   pairs   of   lines   of   the   same   colour   stand   for   the   comparison   between  2  different  radiotherapy  techniques  [73].  ...  13  

Figure  2.8  -­‐  Individually  moulded  thermoplastic  mask  [74].  ...  14  

Figure  2.9  -­‐  Clinical  and  planning  tumour  volumes  (CTV  and  PTV)  [75].  ...  15  

Figure  3.1  -­‐  CT  images  of  different  lung  tumours.  Tumour  contours  are  in  the  left  and  their   respective  three-­‐dimensional  (3D)  visualizations  are  in  the  right  [1].  ...  20  

Figure  3.2  -­‐  Followed  procedure  of  radiomics  [1].  ...  20  

Figure  5.1  -­‐  Kaplan-­‐Meier  Curve  for  the  310  HNSCC  patients  of  the  database.  ...  28  

Figure  5.2  –2D  example  of  a  GLCM  filling  from  a  4x4  image  with  four  grey  levels.  In  this   case,  Ng=4,  δ=1  and  α=0  (horizontal  direction).  ...  31  

Figure  5.3  –  2D  example  of  a  GLRLM  filling  from  a  4x4  image  with  four  grey  levels.  In  this   case,  Ng=4  and  θ=0  (horizontal  direction).  ...  31  

Figure  5.4  -­‐  2D  example  of  a  NGTDM  filling  from  a  5x5  image  with  five  grey  levels.  In  this   case,  d=1,  resulting  in  a  3x3  neighbourhood.  Therefore,  this  neighbourhood  can  only   be  centered  on  the  pixels  within  the  indicated  yellow  square.  The  pixels  outside  this   yellow   square   are   considered   as   being   located   in   the   periphery   of   the   image.   As   an   example  (in  red),  the  first  entry  of  the  matrix   NGTDM(1)  is  given  by   1-­‐!"! ≈ 2.63   since  there  is  only  one  pixel  within  the  yellow  square  with  the  grey  level  1..  ...  32  

Figure  5.5  -­‐  2D  example  of  a  GLSZM  filling  from  a  4x4  image  with  four  grey  levels.  ...  33  

Figure  5.6  -­‐  Transversal,  coronal  and  sagittal  head  and  neck  CT  image  data  visualized  using   CERR  (Computational  Environment  for  Radiotherapy  Research)  [76].  ...  34    

(14)

     

      xiv    

Figure   5.7   –   Transversal   view   of   the   delineation   of   the   GTV-­‐PT   in   the   planning   CT   (left)   and   the   respective   cropped   on   countering   3D   image   (right).   The   features   are   extracted   from   the   image   shown   in   the   right.   All   this   image   data   illustrating   the   strategy  used  for  feature  extraction  was  visualized  using  CERR  [76].  ...  34   Figure  5.8  –  Graph  representation  of  the  stability  rank  for  different  manual  delineations  of   the   tumour   (x-­‐axis)   vs.   the   c-­‐index   obtained   (y-­‐axis)   for   each   one   of   the   440   image   features   implemented   in   previous   work.     A   stability   rank   of   1   indicates   the   most   robust  feature  and  440  the  least  stable  one.  Every  dot  indicates  one  feature  and  the   grey  area  in  the  regression  line  represents  the  confidence  interval  [1].  ...  40   Figure   6.1   -­‐   Common   Kaplan-­‐Meier   curves   for   the   features   Compactness   (2)   and   Sphericity.  ...  47   Figure  6.2  -­‐  Kaplan-­‐Meier  curves  for  the  feature  Grey  level  non-­‐uniformity  mean.  ...  48   Figure  6.3  -­‐  Plot  of  ICC  vs.  c-­‐index  and  respective  regression  line.  Every  circle  indicates  one   feature.   The   green   line   is   the   linear   fit   and   the   broken   red   lines   represent   the   95%   confidence  interval  for  the  prediction  (R!=0.12  and  p-­‐value=6.71E-­‐05).  ...  52   Figure   6.4   –   Pairs   of   Kaplan-­‐Meier   curves   for   the   features   of   the   image   biomarker   signature:  (A)  Uniformity,  (B)  Surface  density  and  (C)  Size  zone  non-­‐uniformity.  ...  55   Figure   6.5   –   Schoenfeld   residuals   method   to   check   the   PH   assumption   for   the   feature/covariate   (A)   Uniformity,   (B)   Surface   density   and   (C)   Size   zone   non-­‐ uniformity.   Plots   of   the   scaled   Schoenfeld   residuals   against   transformed   time   (in   months)  for  each  one  of  the  covariates  in  the  model  fitted  to  the  310  HNSCC  patients.   The  solid  line  is  a  smoothing-­‐spline  fit  to  the  plot  and  the  dashed  lines  represent  a  ±   2-­‐standard-­‐error  band  around  the  fit.  ...  56   Figure  6.6  -­‐  Pre-­‐treatment  risk  stratification  systems  for  OS  based  in  the  image  biomarker   signature  (A)  and  in  the  final  model  (B).  ...  61   Figure   6.7   -­‐   Pre-­‐treatment   risk   stratification   systems   for   RFS   based   in   the   image   biomarker  signature  (A)  and  in  the  final  model  (B).  ...  62                  

(15)

      xv    

List of Tables

Table  6.1  -­‐  Probability  of  survival  after  2  years  of  the  start  of  the  RT  treatment  for  the  310   HNSCC   patients.   These   results   are   shown   according   to   the   values   of   the   3   features  

describing  compactness/sphericity  and  heterogeneity.  ...  48  

Table  6.2  –  Prognostic  performances  obtained  in  the  available  dataset  for  3  image  features   of  the  signature  from  Aerts  et  al.  [1].  ...  49  

Table  6.3  -­‐  Significant  features  of  Group  1  obtained  by  controlling  the  FDR.  ...  49  

Table  6.4  -­‐  Significant  features  of  Group  2  obtained  by  controlling  the  FDR.  ...  50  

Table  6.5  -­‐  Significant  features  of  Group  3  obtained  by  controlling  the  FDR.  ...  50  

Table   6.6   –   The   10   most   prognostic   image   features   and   their   respective   c-­‐index   by   descending  order.  ...  51  

Table  6.7  –  ICC  for  the  10  most  prognostic  image  features.  ...  52  

Table  6.8  -­‐  Image  biomarker  signature  prognostic  performance  and  PH  test  result.  ...  53  

Table   6.9   –   Prognostic   performance   (both   optimistic   and   after   validation)   of   the   multivariable  Cox  model  containing  the  image  biomarker  signature.  ...  54  

Table   6.10   –   DVH   parameters,   their   obtained   p-­‐value   for   the   log   rank   test   and   the   prognostic  performances.  The  variables  that  obtained  significant  survival  differences   are  highlighted  in  bold.  ...  57  

Table  6.11  –  Log  rank  test  result  for  clinical  factors  and  basic  metrics.  The  variables  that   obtained  significant  survival  differences  are  highlighted  in  bold.  ...  58  

Table  6.12  –  Non-­‐validated  prognostic  performances  of  D98  and  the  other  patient-­‐specific   parameters  analysed.  The  variables  that  obtained  significant  survival  differences  are   highlighted  in  bold.  ...  58  

Table   6.13   -­‐   Prognostic   performance   of   the   image   biomarker   signature   in   the   modality   groups.  ...  58  

Table  6.14  -­‐  Prognostic  performance  of  the  image  biomarker  signature  in  the  HPV  status   groups.  ...  58  

Table   6.15–   Complementary   effect   of   the   image   biomarker   signature   in   commonly   used   parameters.  ...  59  

Table  6.16  –  Prognostic  performances  of  patient-­‐specific  variables  in  combination  with  the   image  biomarker  signature.  ...  59  

(16)

     

      xvi    

Table  6.17  –  Log  rank  test  and  probability  of  survival  after  2  years  for  the  pre-­‐treatment   risk  stratification  systems.  ...  60   Table  A.1  -­‐  Image  features  of  Group  1.  ...  II   Table  A.2  –  Image  features  of  Group  2.  ...  IV   Table  A.3    –  Image  features  of  Group  3  extracted  from  the  GLCM.  ...  VII   Table  A.4–  Image  features  of  Group  3  extracted  from  the  GLRLM.  ...  X   Table  A.5  –  Image  features  of  Group  3  extracted  from  the  NGTDM.  ...  XII   Table  A.6  –  Image  features  of  Group  3  extracted  from  the  GLSZM.  ...  XIV                                          

(17)

     xvii  

Table of Contents

CHAPTER  1  INTRODUCTION  ...  1

 

CHAPTER  2  BACKGROUND  ...  3

 

2.1.  HEAD  AND  NECK  CANCER  ...  3  

2.2.  RADIATION  THERAPY  ...  7  

2.3.  RADIOTHERAPY  FOR  HEAD  AND  NECK  CANCER  ...  13  

CHAPTER  3  RADIOMICS  ...  17

 

3.1.  STATE  OF  THE  ART  ...  17  

3.2.  GOAL  OF  PRESENT  STUDY  ...  19  

CHAPTER  4  SURVIVAL  ANALYSIS  ...  21

 

4.1.  KAPLAN-­‐MEIER  METHOD  ...  21  

4.2.  SURVIVAL  MODELLING  ...  23  

CHAPTER  5  PATIENTS  AND  METHODS  ...  27

 

5.1.  PATIENT  DATA  ...  27  

5.2.  IMAGE  FEATURES  ...  28  

5.3.  CT  FEATURE  EXTRACTION  ...  33  

5.4.  DATA  ANALYSIS  ...  35  

CHAPTER  6  RESULTS  ...  47

 

6.1.  VALIDATION  OF  PREVIOUS  PUBLISHED  WORK  ...  47  

6.2.  IMAGE  FEATURES  PROGNOSTIC  POWER  ...  49  

6.3.  MOST  PROGNOSTIC  IMAGE  BIOMARKERS  ...  51  

6.4.  FEATURE  ROBUSTNESS  ...  51  

6.5.  IMAGE  BIOMARKER  SIGNATURE  ...  53  

6.6.  PATIENT-­‐SPECIFIC  PARAMETERS  ...  57  

6.7.  PRE-­‐TREATMENT  RISK  STRATIFICATION  SYSTEMS  ...  60  

CHAPTER  7  DISCUSSION  ...  63

 

7.1.  VALIDATION  OF  PREVIOUS  PUBLISHED  WORK  ...  63  

7.2.  IMAGE  FEATURES  PROGNOSTIC  POWER  ...  65  

7.3.  MOST  PROGNOSTIC  IMAGE  BIOMARKERS  ...  65  

7.4.  FEATURE  ROBUSTNESS  ...  65  

7.5.  IMAGE  BIOMARKER  SIGNATURE  ...  66  

7.6.  PATIENT-­‐SPECIFIC  PARAMETERS  ...  68  

7.7.  PRE-­‐TREATMENT  RISK  STRATIFICATION  SYSTEMS  ...  70  

CHAPTER  8  CONCLUSIONS  AND  FUTURE  WORK  ...  73

 

REFERENCES  ...  75

 

APPENDICES  ...  I

 

A.1.  HISTOGRAM  FEATURES  ...  II   A.2.  GEOMETRIC  FEATURES  ...  IV  

(18)

               

(19)

 

Chapter 1 Introduction

Head   and   neck   cancer   (HNC)   is   ranked   as   the   sixth   most   common   type   of   cancer,   accounting  for  an  estimated  650  000  new  cases  and  350  000  deaths  worldwide,  every   year   [2].   The   most   common   histological   type   of   HNC   is   the   squamous   cell   carcinoma   (SCC),   which   is   usually   detected   in   an   advanced   stage   disease   [3].   In   the   last   two   decades,   an   increase   of   6.5%   (from   27.2%   to   33.7%)   in   the   five-­‐year   overall   survival   (OS)  of  patients  with  SCC  of  the  head  and  neck  has  been  achieved  [4].  This  is  mostly  due   to  recent  developments  of  treatment  options,  such  as  the  administration  of  concomitant   chemotherapy  in  patients  treated  with  radiotherapy  (RT).  However,  survival  rates  still   remain   low   mainly   due   to   loco-­‐regional   recurrences   and   distant   failures   (distant   metastasis),   which   have   poor   treatment   outcomes.   Five-­‐year   local-­‐   and   distant   failure   rates   after   receiving   concomitant   chemoradiotherapy   are   50.8%   and   16.8%,   respectively  [4].      

In  order  to  improve  the  OS  for  HNC  patients  treated  with  RT,  it  would  be  useful  to   know  beforehand,  by  their  tumour  characteristics,  if  a  certain  patient  is  high  or  low  risk.   The  treatment  could  be  planned  accordingly  and,  consequently,  the  outcomes  could  be   much   more   positive.   The   increasingly   tailored   treatment   based   on   specific   characteristics  of  the  patient  and  their  disease  is  called  personalized  medicine  [1].    

Currently,  treatment-­‐  and  patient-­‐specific  parameters,  like  dose-­‐volume  histogram   (DVH)  parameters  and  clinical  tumour  information,  such  as  TNM,  are  used  to  estimate   radiotherapy  treatment  outcome.  However,  based  on  these  parameters  it  is  not  possible   to   obtain   an   accurate   prediction   of   the   probability   of   tumour   control   for   a   specific   patient.  Furthermore,  these  parameters  do  not  provide  a  molecular  characterization  of   the  tumours.    

Molecular   characterization   using   genomics   and   proteomics   technologies   are   already  used  to  characterize  and  classify  tumours.  These  techniques  require  biopsies  or   invasive  surgeries  to  extract  and  analyse  only  small  portions  of  tumour  tissue,  which  do   not   allow   for   a   complete   characterization.   The   applicability   of   these   procedures   is   limited   because   tumours   are   spatially   and   temporally   heterogeneous.   Furthermore,   genomics   and   proteomics   are   still   challenging   techniques   to   implement   into   clinical   routine.    

In   this   context,   medical   imaging   arises   as   a   good   option   as   is   routinely   used   worldwide   in   clinical   practice,   for   oncologic   diagnosis   and   treatment   guidance.   In   addition,   imaging   is   a   non-­‐invasive   procedure   and   can   provide   a   more   comprehensive  

(20)

  2   view   of   the   entire   tumour,   having   great   potential   to   monitor   the   development   and   progression   of   the   disease   or   its   response   to   therapy.   The   existing   modalities   in   oncology,   such   as   Magnetic   Resonance   Imaging   (MRI),   Positron   Emission   Tomography   (PET)   and   Computed   Tomography   (CT),   can   exhibit   strong   phenotypic   differences   in   tumours.   The   most   widely   used   imaging   modality   in   oncology   is   X-­‐Ray   CT,   which   evaluates  tissue  density.  This  technique  will  be  the  one  used  for  this  project.  

The   hypothesis   for   this   project   is   that   different   tumour   phenotypes,   that   can   be   detected  by  image  modalities,  relate  to  patient-­‐specific  response  to  RT.  The  process  that   allows  this  extraction  is  called  radiomics.    

Radiomics   is   an   emerging   field   able   to   characterize   an   image   quantitatively.   Therefore,   the   ultimate   aim   of   this   project   is   to   predict   RT   treatment   outcome   of   a   specific  patient  based  on  their  quantitative  tumour  CT  image  characteristics.  

Briefly,   the   remainder   of   the   thesis   is   organized   as   follows:   Chapter   2   provides   a   background  for  HNC  and  explains  in  detail  the  followed  steps  in  the  University  Medical   Center   Groningen   (UMCG)   for   the   treatment   of   HNC   through   RT;   In   Chapter   3,   the   concepts  of  radiomics  are  introduced,  followed  by  the  state  of  the  art  and,  subsequently,   a   description   of   the   main   contributions   of   the   present   study   in   the   field;   Chapter   4   exposes   some   definitions   regarding   survival   analysis   that   need   to   be   retained   for   the   understanding   of   the   rest   of   the   thesis;   In   Chapter   5   the   detailed   description   of   the   patients  used  for  this  study  and  the  methods  of  both  image  processing  and  data  analysis   are  explained;  Chapter  6  shows  all  the  results  obtained  and  Chapter  7  their  respective   discussion;  Finally,  Chapter  8  gives  a  conclusion  of  all  the  work  performed,  mentioning   some   current   limitations   of   this   study   and   also   interesting   topics   to   develop   as   future   work.                          

(21)

Chapter 2 Background

An   overview   of   head   and   neck   cancer   (HNC),   some   important   aspects   of   radiotherapy,   and   also   the   treatment   of   HNC   through   radiotherapy   are   given   in   this   chapter  for  the  comprehension  of  the  aim  of  this  project.  

 

2.1. Head and Neck Cancer

Head  and   neck  cancer  (HNC)  represents  about  6%  of  all  diagnosed  cancer   cases   [2].  The  median  age  for  diagnosis  of  HNC  patients  is  the  early  sixties  and  it  is  a  type  of   malignancy  more  common  in  male  patients.    

Cigarette   smoking,   the   use   of   smokeless   tobacco   and   also   alcohol   abuse   have   shown   to   be   the   major   environmental   risk   factors   for   HNC   [5].   In   addition,   Human   Papillomavirus  (HPV)  was  suggested  to  be  of  influence  in  the  development  of  HNC  [6].   Common   symptoms   of   HNCs   consist   of   resistant   mucosal   ulcers,   sore   throat,   referred   otalgia,  dysphagia,  chronic  cough  and  neck  mass.  

 

2.1.1. Diagnosis

For   the   classification   and   attribution   of   stages   for   HNC,   the   diagnosis   of   the   tumour   has   to   be   concluded.   To   achieve   a   definitive   diagnosis,   physical   examination,   blood   tests   and   histological   biopsy   can   be   performed.   In   addition,   imaging   modalities,   such   as   X-­‐rays,   Computed   Tomography   (CT),   Magnetic   Resonance   Imaging   (MRI),   Ultrasound  and  18F-­‐Fluordesoxyglucose  Positron  Emission  Tomography  (FDG-­‐PET)  can   be  used.  Particularly,  ultrasound  complemented  with  fine  needle  aspiration,  provide  the   extraction  and  pathological  evaluation  of  a  portion  of  the  tumour.    

In   addition,   for   examination   of   mucosa   in   the   upper   aerodigestive   tract,   a   nasopharyngolaryngoscope   is   used.   By   displacing   this   flexible   fibre-­‐optic   endoscope   through   the   naso-­‐,   oro-­‐,   hypopharynx   and   superior   region   of   the   larynx,   optical   information   on   location   and   superficial   local   invasion   of   the   malignant   tumour   can   be   acquired.      

(22)

  4  

2.1.1.1. Classification

 

Cancers  can  be  classified  in  two  ways:  by  the  type  of  tissue  in  which  the  cancer   originates   (histological   type)   and   by   primary   site,   which   is   the   location   in   the   body   where  the  cancer  was  first  developed.    

The  primary  site  of  HNCs,  as  the  name  implies,  is  in  the  head  and  neck  region.   Despite  a  small  number  that  arises  in  the  paranasal  sinuses,  thyroid  and  salivary  glands,   most   of   HNCs   arise   from   the   upper   aerodigestive   tract,   which   is   composed   by   four   anatomical  subsites:  oral  cavity,  oropharynx,  hypopharynx  and  larynx  (Figure  2.1).    

                             

More  precisely,  the  oral  cavity  is  defined  anteriorly  by  the  lips,  superiorly  by  the   hard   and   soft   palate   junction   and   inferiorly   by   the   circumvallate   papillae.   The   oropharynx   is   defined   anteriorly   by   the   circumvallate   papillae   and   posteriorly   by   the   posterior   pharyngeal   wall.   The   hypopharynx   is   superiorly   bounded   by   the   hyoid   bone   and   extends   to   the   inferior   border   of   the   cricoid   cartilage.   Finally,   the   larynx   extends   from  the  tip  of  the  epiglottis  to  the  inferior  edge  of  the  cricoid  cartilage.  

Regarding   the   classification   of   cancers   based   on   histological   type,   there   are   hundreds   of   different   cancers,   which   can   be   grouped   into   the   following   categories:   carcinoma,   sarcoma,   myeloma,   leukaemia,   lymphoma,   blastoma   and   mixed   types.   However,   carcinoma   is   the   most   common   histological   type   and   it   consists   in   malignancies  derived  from  epithelial  cells.  Epithelial  tissue  is  present  in  the  skin,  as  well  

Figure  2.1  -­‐  The  anatomical  subsites  of  the  upper  aerodigestive  tract  [70].  

(23)

as   in   the   covering   and   lining   of   organs   and   internal   passageways,   such   as   the   gastrointestinal   tract.   Carcinomas   are   divided   into   two   major   subtypes:   adenocarcinoma,   which   develops   in   an   organ   or   gland,   and   squamous   cell   carcinoma   (SCC),   which   originates   in   the   squamous   epithelium.   Additionally,   adenocarcinomas   generally  occur  in  mucus  membranes  and  SCCs  can  appear  in  many  areas  of  the  body.    

Since  all  of  the  mentioned  subsites  of  the  upper  aerodigestive  tract  are  lined  by   squamous  mucosa,  the  most  common  histological  type  of  malignancies  in  these  regions   is   the   carcinoma,   particularly,   the   SCC   (Figure   2.2).   In   fact,   over   90%   of   all   upper   aerodigestive  tract  malignancies  are  SCCs  [5].  In  the  stage  of  presentation,  these  types  of   malignancies   usually   involve   regional   lymph   nodes   but   the   presence   of   distant   metastasis  is  uncommon.    

                          2.1.1.2. TNM Staging  

The   TNM   (Tumour,   Node,   Metastasis)   is   a   cancer   staging   notation   system,   developed  and  maintained  by  the  Union  for  International  Cancer  Control  (UICC),  which   is   currently   used   in   the   oncology   field   for   risk   stratification   and   treatment   decision-­‐ making.    

With   this   classification   system,   consensus   on   one   globally   recognised   standard   for  the  classification  of  the  extent  of  spread  of  cancer  of  a  patient  is  achieved.  Since  this   notation  is  only  used  for  malignancies  originated  from  a  solid  tumour,  it  is  applicable  for   HNC,  in  which  each  type  has  its  own  TNM  classification.    

Figure  2.2  -­‐  Endoscopic  examination  of  a  SCC  of  the  larynx  [71].    

(24)

  6   Adopting  TNM  brings  several  benefits  to  the  clinical  field,  such  as  helping  to  plan   the  treatment  and  giving  an  indication  of  its  prognosis.  In  addition,  the  TNM  system  can   assist  in  the  evaluation  of  the  results  of  treatment  and  also  enables  facilities  around  the   world  to  interchange  information  more  productively  [7].  

In   this   notation,   T   describes   the   size   of   the   original   (primary)   tumour   and   whether  it  has  invaded  nearby  tissue.  N  describes  nearby  (regional)  lymph  nodes  that   are  involved.  Finally,  M  describes  distant  metastasis  (spread  of  cancer  from  one  part  of   the  body  to  another).  After  obtaining  a  value  for  each  one  of  these  three  parameters,  it  is   possible   to   group   specific   values   of   the   three   parameters   to   define   different   stages,   which  are  associated  with  prognosis  (Figure  2.3).    

 

 

2.1.2. Treatment

The   predominant   treatment   modalities   for   HNC   patients   consist   of   surgery,   radiation  therapy  (RT)  and  chemotherapy,  individually  or  in  combination.  The  selection   and   planning   of   the   most   appropriate   treatment   for   head   and   neck   squamous   cell   carcinoma   (HNSCC)   is   often   complicated   and   can   involve   an   evaluation   by   a   multidisciplinary   team,   including   head   and   neck   surgeons,   medical   oncologists,   dieticians,  radiation  oncologists,  radiologists,  plastic  surgeons  and  dentists.  In  addition,   primary  tumour  site,  stage,  resectability  and  also  patient  factors  (as  swallowing,  airway   considerations,  desire  for  organ  preservation  and  illness)  can  interfere  in  the  treatment   selection  [2].  

Oral   cavity   and   oropharyngeal   SCCs   are   the   most   common   types   of   HNC.   In   general,  early  stage  lesions  can  be  treated  with  either  RT  or  surgery,  with  radiotherapy   generally   yielding   better   functional   outcomes   [8].   However,   stages   III   and   IV   cancers,  

(25)

with  worse  prognosis,  are  treated  with  surgery  followed  by  RT.  In  case  of  unresectable   lesions   in   these   late   stages,   the   cancer   is   treated   with   combined   radiation   and   chemotherapy.  

The   treatment   of   early   stages   of   hypopharyngeal   cancers   can   be   accomplished   through  either  surgery  or  RT.  In  contrast,  higher  stage  lesions  require  surgery  followed   by   RT.   The   spread   of   this   type   of   cancer   to   regional   lymph   nodes   (loco-­‐regional   metastasis)  is  common  and  must  be  addressed  with  neck  dissection  and/or  RT  [5].    

Early   laryngeal   cancer   can   be   treated   with   RT   or   surgery.   More   advanced   lesions   can  be  treated  with  either  surgery  followed  by  RT  or  combined  RT  and  chemotherapy.   In   all   of   these   procedures,   a   considerable   effort   is   performed   in   order   to   spare   the   larynx.  

 

2.2. Radiation Therapy

 

2.2.1. External Beam Radiotherapy Modalities

Intensity-­‐Modulated   Radiation   Therapy   (IMRT)   and   Volumetric   Modulated   Arc   Therapy  (VMAT)  are  two  modalities  for  external  beam  RT  that  arise  as  alternatives  to   the  former  standard  modality  Three-­‐Dimensional  Conformal  Radiotherapy  (3D-­‐CRT).    

2.2.1.1. Conformal Radiation Therapy

 

In  Three-­‐Dimensional  Conformal  Radiotherapy  (3D-­‐CRT),  precise  tumour  definition   and   more   accurate   dose   calculations   (by   accounting   for   axial   anatomy   and   complex   tissue  contours)  are  achieved  by  the  use  of  Computed  Tomography  (CT)  images  in  the   planning   of   the   treatment   [9].   Additionally,   in   conventional   3D-­‐CRT,   delivery   patterns   (beam  arrangement,  filtration  and  shaping)  are  introduced  into  the  treatment  planning   system  in  order  to  deliver  an  adequate  radiation  treatment  to  the  patient.    

With  this  delivery  technique,  it  is  possible  to  achieve  suitable  target  coverage,  while   sparing  healthy  tissue,  by  the  variation  of  the  number  and  direction  of  fields  and  weight   of   the   beams.   Additionally,   3D-­‐CRT   can   be   delivered   due   to   the   emergence   of   certain   components   in   the   treatment   head   of   the   linear   accelerator   (LINAC),   which   will   be   subsequently  explained  in  more  detail  in  this  text.  For  the  photon  mode  therapy,  wedge  

(26)

  8   filters   and   custom   blocks   or   multileaf   collimators   (MLCs)   can   be   used.   Conversely,   for   electron  therapy,  applicators  are  inserted  instead.    

Custom  blocks  are  thick  pieces  of  lead  that  are  used  to  shape  the  field.  They  can  be   loaded   into   the   accessory   tray   as   an   additional   means   of   collimating   the   beam.   These   blocks   are   specifically   for   an   individual   patient   to   shield   sensitive   tissue   or   structures   and  can  be  made  to  the  exact  shape.  However,  custom  blocks  have  fallen  out  of  use  due   to  multileaf  collimation,  which  is  faster,  can  provide  more  flexibility  and  is  less  taxing  on   staff.    

MLCs   have   been   introduced   for   use   in   radiotherapy   in   1980   and   are   now   widely   used,   having   become   an   integral   part   of   any   radiotherapy   department   [10].   These   devices  have  up  to  80  pairs  of  individual  leaves  made  of  high  atomic  numbered  material,   usually  tungsten  and  can  move  independently  in  and  out  of  the  path  of  a  particle  beam   in  order  to  form  the  desired  field  shape  (Figure  2.4).  

There  are  frequent  occasions  when,  instead  of  a  uniform  intensity  across  the  beam   profile,  a  wedge-­‐shaped  intensity  profile  is  required  for  the  photon  beam.  The  wedge  is   used  to  compensate  for  different  attenuations  caused  by  the  tissues,  thus  being  possible   to  obtain  a  homogeneous  dose  distribution.  To  this  end,  the  wedge  angle  is  previously   defined,   and   this   is   the   angle   from   the   perpendicular   line   of   the   central   axis   to   the   isodose  curve  within  the  treated  volume.  

Finally,   applicators   are   used   to   provide   additional   collimation   with   the   aim   of   keeping   a   sharp   electron   beam   edge   within   the   treatment   volume   [10].   The   lower   sections  are  made  up  of  successive  collimation  layers  that  absorb  electrons  that  would   scatter  out  of  the  field.  In  addition,  the  applicator  typically  has  an  additional  tray  for  the   insertion   of   electron   cut-­‐outs.   Unlike   the   custom   blocks   for   photon   treatments,   the   electron  cut-­‐outs  are  much  thinner  and  lie  very  close  to  the  patient’s  surface  in  order  to   reduce  the  size  of  the  side  of  the  beam  (or  penumbra).    

(27)

               

2.2.1.2. Intensity-Modulated Radiation Therapy

 

Since   2006,   Intensity-­‐Modulated   Radiation   Therapy   (IMRT)   replaced   3D-­‐CRT   for   the  treatment  of  the  majority  of  HNC  patients  in  the  UMCG.  IMRT  has  been  considered   the   most   exciting   development   in   RT   since   the   introduction   of   3D   imaging   (CT)   to   the   treatment   planning   process   [11].   Besides   IMRT   being   an   advanced   form   of   3D-­‐CRT,   these   two   techniques   differ   in   the   procedure   for   treatment   planning   and   treatment   execution  at  the  LINAC.    

Fluence  can  be  seen  as  the  amount  of  radiation  that  comes  out  of  the  LINAC  in  a   certain   direction.   The   great   innovation   that   arises   with   IMRT   is   the   possibility   of   creation   of   multiple   segments   in   one   beam   by   the   MLC,   which   leads   to   a   non-­‐uniform   radiation   fluence   in   each   field   direction   (modulated   intensities).   Therefore,   the   superposition   of   modulated   beams   allows   the   delivery   of   a   highly   conformal   dose   distribution  to  the  patient,  in  which  high  doses  are  addressed  to  tumour  (target)  tissues   (Figure   2.5).   Simultaneously,   since   concavities   in   the   high-­‐dose   irradiated   volume   are   present,   this   type   of   radiotherapy   allows   the   sparing   of   healthy   and   critical   structures   adjacent  to  the  tumour,  such  as  the  parotid  glands,  spinal  cord  and  brainstem  [12].  The   side   effects   caused   by   the   radiation   delivered   with   IMRT   are   generally   reduced   in   comparison   to   3D-­‐CRT   [13].   In   addition,   IMRT   has   shown   to   be   effective   for   the   loco-­‐ regional  control  of  HNC  [14].  

Figure  2.4  -­‐  An  MLC  perpendicular  to  the  eye  view  of  the  beam  [11].  

(28)

  10   However,   IMRT   requires   more   time   for   the   delivery   of   the   treatment   and   an   increased   number   of   monitor   units   (MU)   -­‐   the   LINAC   output   -­‐   for   a   given   fraction,   resulting   in   a   greater   integral   body   dose,   which   may   lead   to   a   higher   risk   for   the   development  of  second  malignancies  [15],  [9].  

The   modulation   of   fluence   caused   by   the   MLC   in   IMRT   can   be   achieved   with   dynamic   or   static   field   segments   (step   and   shoot).   In   the   step   and   shoot   technique,   whenever   the   MLC   is   changing   its   position,   the   irradiation   stops   and   starts   in   the   following   arrangement   of   the   leaves.   Conversely,   in   dynamic   IMRT   the   radiation   is   delivered  while  the  leaves  move  to  create  the  next  segment  [9].  

Unlike   3D-­‐CRT,   in   IMRT,   the   specific   dose   and   volume   constraints   previously   defined   by   the   clinician   are   used   as   an   input   and,   subsequently,   an   individualized   optimal  treatment  plan  that  complies  these  requirements  is  developed  (so  called  inverse   treatment   planning)   [16].   However,   the   resulting   optimal   plan   is   impaired   by   known   limitations  of  the  treatment  unit,  such  as  physical  and  mechanical  characteristics  of  the   MLC   (leakage   and   gaps   between   the   leaves).   Therefore,   the   computer-­‐based   optimization   techniques   used   in   IMRT   planning   are   based   on   finding   the   global   minimum   of   the   cost   function,   which   can   be   represented   in   each   interaction   point   (or   fluence)  by:  

 

 

𝑐𝑜𝑠𝑡 = 𝐼(𝑥, 𝑦, 𝑧) 𝐷 𝑥, 𝑦, 𝑧 − 𝐷

!

(𝑥, 𝑦, 𝑧)

!,   Eq.  2.1  

 

where   𝐼(𝑥, 𝑦, 𝑧)   is   the   importance   of   the   voxel   defined   by   (𝑥, 𝑦, 𝑧),   𝐷!(𝑥, 𝑦, 𝑧)   is   the   required  prescribed  dose  distribution  and  𝐷 𝑥, 𝑦, 𝑧  the  actual  delivered  dose  [12].  The   importance   𝐼(𝑥, 𝑦, 𝑧)  of  the  cost  function  can  be  varied  by  the  planner  according  to  the   pathology  of  the  patient,  by  giving  priority  in  complying  the  constraints  in  the  voxels  of   the  target  or  healthy  tissues,  or  even  creating  a  compromise  between  the  two.    

Volumetric   Modulated   Arc   Therapy   (VMAT)   is   a   novel   RT   modality,   which   is   introduced   recently   for   the   treatment   of   HNC   patients   in   the   UMCG.   This   technique   delivers  radiation  dose  (with  constant  or  variable  rate)  accurately  and  efficiently  using  a   360°   range   of   directions   in   a   single   gantry   arc   [15].   In   comparison   to   IMRT,   VMAT   includes   a   larger   number   of   beam   directions,   providing   a   much   faster   treatment   with   reduction  in  the  number  of  MU  required  [17],  [18].    

However,  this  technique  has  the  drawback  of  requiring  additional  constraints  on   the   MLC   leaf   motions   because   of   the   continuous   motion   of   the   gantry   throughout   delivery  of  the  treatment  [19].  This  is  one  of  the  reasons  why  VMAT  is  still  not  as  widely   implemented  as  IMRT  in  the  clinical  RT  departments.  

Referências

Documentos relacionados

decrease as compared to Ti, reaching an ~80% reduction when a longer spacer was used (Ch_AHA_Nt-Dhvar5 and Ch_GG_Nt- Dhvar5) (p < 0.05). These two Dhvar-bearing surfaces did

Ousasse apontar algumas hipóteses para a solução desse problema público a partir do exposto dos autores usados como base para fundamentação teórica, da análise dos dados

3 Specialist in Head and Neck Surgery, Peroral Endoscopy, and Otorhinolaryngology (Assistant Professor in the Department of Surgery of the Medical School of the Federal University

5 PhD in Pathology from the Medical School of the University of São Paulo (Head of the Department of Head and Neck Surgery and Otorhinolaryngology at the Heliópolis Hospital,

Effect of low level helium- neon (He-Ne) laser therapy in the prevention and treatment of radiation induced mucositis in head and neck cancer patients.

The probability of attending school four our group of interest in this region increased by 6.5 percentage points after the expansion of the Bolsa Família program in 2007 and

Diante do exposto, o presente estudo mostra-se relevante no campo da saúde pública no sentido de reforçar os achados disponíveis, ampliar as evidências sobre a relação

ESTRUTURA E SIMILARIDADE FLORÍSTICA DE DOIS COMPONENTES ARBÓREOS DE FLORESTAS ESTACIONAIS SEMIDECIDUAIS DO PARQUE ESTADUAL DAS VÁRZEAS DO RIO IVINHEMA-MS Shaline Sefara Lopes