utd_medknow
Indian Journal of Cancer
Home  ICS  Feedback Subscribe Top cited articles Login 
Users Online :263
Small font sizeDefault font sizeIncrease font size
Navigate here
  Search
 
  Ahead of print
  
Resource links
    Search Pubmed for
 
    -  Akcay M
    -  Etiz D
    -  Celik O
    -  Ozen A

 
  In this article
   Abstract
  Introduction
   Materials and Me...
  Results
  Discussion
  Conclusion
   References
   Article Figures
   Article Tables

 Article Access Statistics
    Viewed137    
    PDF Downloaded10    

Recommend this journal

 

Previous Article  Table of Contents  Next Article
ORIGINAL ARTICLE
Ahead of print publication
 

Evaluation of acute hematological toxicity by machine learning in gynecologic cancers using postoperative radiotherapy


1 Department of Radiation Oncology, Medical Faculty of Osmangazi University, Eskişehir, 26480, Turkey
2 Department of Mathematics - Computer Eskisehir Osmangazi University, Eskişehir, 26480, Turkey

Date of Submission24-Jul-2019
Date of Decision20-Jun-2020
Date of Acceptance07-Oct-2020
Date of Web Publication02-Jul-2021

Correspondence Address:
Melek Akcay,
Department of Radiation Oncology, Medical Faculty of Osmangazi University, Eskişehir, 26480
Turkey
Login to access the Email id

Source of Support: None, Conflict of Interest: None

DOI: 10.4103/ijc.IJC_666_19

  Abstract 


Background: The aim of the study is to investigate the factors affecting acute hematologic toxicity (HT) in the adjuvant radiotherapy (RT) of gynecologic cancers by machine learning.
Methods: Between January 2015 and September 2018, 121 patients with endometrium and cervical cancer who underwent adjuvant RT with volumetric-modulated arc therapy (VMAT) were evaluated. The relationship between patient and treatment characteristics and acute HT was investigated using machine learning techniques, namely Logistic Regression, XGBoost, Artificial Neural Network, Random Forest, Naive Bayes, Support Vector Machine (SVM), and Gaussian Naive Bayes (GaussianNB) algorithms.
Results: No HT was observed in 11 cases (9.1%) and at least one grade of HT was observed in 110 cases. There were 55 (45.5%) cases with ≤grade 2 HT (mild HT) and 66 (54.5%) cases with grade ≥3 HT (severe HT). None of the patients developed grade 5 HT. Of 24 variables that could affect acute HT, nine were determined as important variables. According to the results, the best machine learning technique for acute HT estimation was SVM (accuracy 70%, area under curve (AUC): 0.65, sensitivity 71.4%, specificity 66.6%). Parameters affecting hematologic toxicity were evaluated also by classical statistical methods and there was a statistically significant relationship between age, RT, and bone marrow (BM) maximum dose.
Conclusion: It is important to predict the patients who will develop acute HT in order to minimize the side effects of treatment. If these cases can be identified in advance, toxicity rates can be reduced by taking necessary precautions. These cases can be predicted with machine learning algorithms.


Keywords: Cancer, machine learning, radiotherapy, toxicity
Key Message : Myelosuppressive effects of radiotherapy are known. It is important to predict the patients who will develop acute hematological toxicty in order to minimize the side effects. These cases can be predicted with machine learning algorithms.



How to cite this URL:
Akcay M, Etiz D, Celik O, Ozen A. Evaluation of acute hematological toxicity by machine learning in gynecologic cancers using postoperative radiotherapy. Indian J Cancer [Epub ahead of print] [cited 2021 Jul 27]. Available from: https://www.indianjcancer.com/preprintarticle.asp?id=320442





  Introduction Top


Myelosuppressive effects of radiotherapy (RT) and chemotherapy (ChT) are known. Radiation causes myelosuppression and characteristic pathological and radiographic changes of bone marrow (BM), leading to apoptosis of the BM stem cells and stromal damage.[1] BM deficiency due to BM toxicity is defined as the inability to produce appropriate blood cells required to control immune function.[2] There are two different types of stem/progenitor cells in BM. With respect to hematopoiesis, hematopoietic stem/progenitor cells produce mature blood cells, and this process is supported by mesenchymal stromal stem cells.[3]

Pelvic BM is the main site of hematopoiesis. In standard pelvic irradiation, a significant amount of BM is involved in the radiotherapy field, and this affects hematopoietic stem cells.[4] Pelvic RT plays an important role in the treatment of gynecologic cancers, especially cervix and endometrial cancer. More than 50% of the BM of the body is found in os coxae, sacrum, proximal femur, and lower lumbar spine.[5] A standard pelvic RT involves approximately 40% of the total BM and may result in hematologic toxicity (HT).[6] BM activity decreases after RT, with the BM regeneration varying according to the radiation dose used.[7]

Pelvic RT was traditionally performed with a three-dimensional conformal RT (3DCRT) using the four-field box technique based on standard anatomical boundaries. After surgery, a part of the gastrointestinal tract changes toward the pelvic space and leads to clinically significant toxicity. At the same time, a significant portion of the BM reserve is in the RT field, and hematopoietic stem cell damage and BM toxicity occur when 3DCRT is used. Intensity-modulated RT (IMRT) has been investigated in many studies as a technique to reduce the doses to organs-at-risk (OAR), while maintaining the dose delivered to the target volume, and statistically significant decreases have been found in the small bowel, bladder, rectum, and BM volume.[8],[9] There are many studies evaluating HT secondary to RT in gynecologic cancer.[10],[11],[12] In these studies, 3DCRT and IMRT techniques were evaluated, but in the current study, we used the volumetric-modulated arc therapy (VMAT) as a new technique.

In radiation oncology, there are typical clinical questions, such as “Which patients have the highest risk of toxicity?” and “What is the probability of local control success?”. Although there are gold-standard clinical studies to answer these questions, these studies are costly, long-lasting, and limited in the number of feasible patients. Using the available data, future clinical trials can be better planned and new findings can be obtained. Evidence-based medicine is based on randomized controlled trials designed with a large patient population. However, the number of clinical and biological parameters that should be investigated in order to obtain sensitive results is increasing every day.[13] A new methodology emerges from routine data to support clinical decisions.[14] Lambin et al. have described in detail the following categories of features that should be considered and integrated into a prediction model[15]:

  1. Clinical (Patient performance, tumor grade, blood test results, and patient questionnaires)
  2. Treatment (Dose distribution, ChT)
  3. Imaging (Tumor size and volume, metabolic uptake, radiomics)
  4. Molecular (intrinsic radiosensitivity, hypoxia, genomic studies).


Prediction-based modeling is a two-stage process that involves qualification and approval. Qualification refers to the process, in which the data is shown to be an indicator of the result. Once predictive or prognostic factors have been identified, they must be verified using a different set of data. After these factors have been qualified and approved, further studies should be carried out to assess whether model-based decisions actually improve the outcomes of a model.

Machine learning algorithms (for example, survival, response to treatment, relapse, toxicity) have been attracting attention recently for predicting the results of RT and ChT. Estimating toxicity is important in improving treatment and providing information to patients and clinicians in cases treated for cancer. Considering the cases diagnosed with gynecological cancer receiving radiotherapy with a specific demographic, tumor, and treatment information, it is an important issue whether the development of toxicity can be estimated by any parameter.

The aim of this study was to evaluate the factors affecting HT in patients diagnosed with endometrium and cervical cancer treated with adjuvant pelvic RT with the VMAT technique using machine [Median (range)] learning.


  Materials and Methods Top


Patients

The study included 121 patients with gynecologic cancer who received an adjuvant pelvic RT in the Radiation Oncology Department of Eskişehir Osmangazi University, Faculty of Medicine, between January 2015 and September 2018. This study was approved by the Ethics Committee of the Human Studies Review Board of Eskisehir Osmangazi University.

The inclusion criteria were having stage I-III endometrium/cervical cancer and having undergone a postoperative pelvic RT with the VMAT technique. Patients with a history of pelvic RT, Karnofsky Performance Score (KPS) <70, and distant metastasis were not included in the study. During RT, the patients were evaluated at least once a week by a physician, and a weekly complete blood count was undertaken. Acute HT assessment was performed according to the Common Terminology Criteria for Adverse Events (version 5.0). The toxicity seen within ≤90 days from the start of RT was accepted as acute toxicity.

Treatment plan

The patients were immobilized in a supine position. All procedures were planned using a computed tomography (CT)-based simulation. The CT slices were 5 mm thick for all patients and the scan was performed with a Siemens Somatom Definition AS® CT device. Each patient underwent the procedure with the urinary bladder full. For all patients included in the study, all pelvic bones were contoured 1 cm above and below the upper and lower limit of the planning target volume (PTV). Contouring was performed by two dosimetrists. The contours were constructed to include the femoral heads, sacrum, and lumbar vertebrae. Lymphatics were contoured according to the Radiation Therapy Oncology Group (RTOG) guide.[16] PTV was created by adding a 0.7 mm margin to the clinical target volume (CTV), and 45–54 Gy was planned to be delivered to PTV. The OAR doses were V40 <30% for the small bowel, V30 <60% for the rectum, V45 <35% for the bladder, and V30 <15% for the femoral head. Full bladder protocol was used in CT simulation and treatment. The internal target volume was not used. These plans were constructed by two physicists and calculated using the (Analytical Anisotropic Algorithm) AAA in the Eclipse planning system (Version 13.0.26, Varian Medical Systems, Palo Alto, CA, USA). The patients were treated with a Varian Trilogy® linear accelerator.

Kilovoltage port images were performed before each treatment and cone-beam CT (CBCT) imaging was performed on the first 3 days of treatment and then once a week.

Evaluation of data—Machine learning

The patient characteristics given in [Table 1] were analyzed by correlation analysis. The normality test, correlation analysis, and binary logistic regression analysis were used to determine the variables with a significant effect on mild and severe HT. Using these methods, the following nine variables were selected from a total of 24 variables: age, KPS, diagnosis (endometrium/cervical cancer), disease stage, presence of metastatic lymph nodes, histopathology, RT dose, history of ChT, BM volume, BM V5-10-20-30-40-45 (%), BM V5-10-20-30-40-45 (volume), and the minimum, mean, maximum doses of BM. The key features selected from these characteristics are summarized in [Table 2]. The effect factors of the determining variables are given in [Figure 1]. A binary logistic regression model was constructed using the nine independent variables on the data set. For machine learning, 70% of the data (belonging to 84 patients) was used for training and the remaining 30% (belonging to 37 patients) for testing. After the training of the data, a prediction model was formed. Acute HT estimation was performed using different machine learning algorithms with the help of the model. For this estimation, Logistic Regression, XGBoost, Artificial Neural Network (ANN), Random Forest, Naive Bayes, Support Vector Machine (SVM), and Gaussian Naive Bayes (GNB) machine learning algorithms were used.
Table 1: Patient and treatment characteristics

Click here to view
Table 2: Variables affecting hematological toxicity

Click here to view
Figure 1: The effect factors of the variables

Click here to view


In the current study, a commonly used success evaluation method, accuracy rate, was utilized. The accuracy method involves the calculation of the ratio of a system's correctly classified instances (true positive [TP], and true negative [TN]) to the total number of instances. The error rate refers to the ratio of the number of incorrectly calculated instances (false positive [FP] and false negative [FN]) to the total number of instances. The success rates calculated using the confusion matrix. Below are the success criteria and formulas used in the current study and calculated based on the confusion matrix.[17]

Accuracy = (TP + TN) / (TP + TN + FN + FP)

Sensitivity = TP/(TP + FP)

Negative Prediction Value = TN/(TP + FP)

Recall = TP/(TP + FN)

Specificity = TN/(FP + TN)

Statistics

SPSS for Windows 21.0 (IBM Corp. Released 2013. IBM SPSS Statistics for Windows, Version 22.0. Armonk, NY: IBM Corp.) was used in the analysis. For toxicity analyzes; Chi-square, t-test, and Mann-Whitney U test tests were used. Data are summarized as mean. P < 0.05 was considered statistically significant.

Among the groups with and without HT, stepwise binary logistic regression (Backward Wald) was performed and no statistically significant variable was found in the statistical model.


  Results Top


A total of 103 (85.1%) endometrium and 18 (14.9%) cervical cancer cases were included in the study. The median age was 62 (range: 31–88) years. The patient and treatment characteristics are summarized in Table 1. Only one of the cases with cervical cancer received concomitant chemotherapy with RT. All the remaining 26 cases that underwent chemotherapy were diagnosed with endometrial cancer and received consecutive chemotherapy with RT. The median follow-up period was 16 months. The median volume of BM was 1199.8 (range: 565.5–1598.5) cc. The median BM V5, V10, V20, V40, and V45 were 92.6%, 83.7%, 52.7%, 17.8%, 5.5%, and 0.8%, respectively. The dosimetric parameters are summarized in [Table 3]. The treatment plan and the dose-volume histogram (DVH) examples are given in [Figures 2a and b].
Table 3: Dosimetric parameters

Click here to view
{Figures 2a and b}

Hematologic toxicity

No HT was observed in 11 cases (9.1%) while the remaining 110 cases presented with HT, of whom 66 (54.5%) had grade 0-2 HT and 55 (45.5%) had grade 3-4. None of the patients developed grade 5 HT. HT was most observed in lymphocytes and least in platelets. Hemoglobin, leukocyte, neutrophil, lymphocyte, and platelet toxicities are summarized in [Table 4].

After the training of the data, a prediction model was formed. BM toxicity was estimated using different machine learning algorithms based on age, KPS, metastatic lymph node presence, RT dose, ChT history, BM volume, BM V5, and BM minimum and maximum dose data. The accuracy rates of the results are given in [Table 5].
Table 4: Hematological toxicity (HT) by cell type

Click here to view
Table 5: The results of the estimation of BM toxicity obtained by different machine learning algorithms

Click here to view


In the common iliac and paraaortic lymph node involvement, BM volume that is included in the treatment field increases due to the application of extended field RT. It is expected that the risk and rate of toxicity will increase due to the increase in BM volume entering the treatment field.

The Random Forest and SVM algorithms were found to be the best machine learning techniques for acute HT estimation according to the accuracy rates. In the prediction of acute HT, according to the area under the curve (AUC) values, SVM (AUC = 0.65) was superior to Random Forest (AUC = 0.52). SVM had a confidence interval (CI) of 0.55–0.85, sensitivity of 71.4%, and specificity of 66.6%. The AUC index chart for SVM is shown in [Figure 3]. The confusion matrix revealed that SVM accurately predicted six of 14 severe HT cases, indicating an accurate prediction rate of 43% for these cases. According to the same matrix, 20 of 23 patients with a history of mild HT in the test data were accurately predicted by SVM, and the accurate estimation success was 87% [Table 6]. Thus, the success of the SVM technique in predicting acute HT patients was found to be 70%.
Figure 3: SVM AUC index; SVM: Support Vector Machine; AUC: Area Under Curve

Click here to view
Table 6: Confusion matrix (SVM)

Click here to view


Parameters affecting hematologic toxicity were evaluated by also classical statistical methods. There was a statistically significant relationship between leukocyte toxicity and age. The median age was 60 in cases with leukocyte toxicity and 63 in cases without leukocyte toxicity (P = 0.023). A statistically significant relationship was found between hemoglobin toxicity and RT total dose (P = 0.014) and BM maximum dose (P = 0.042). In patients with anemia, the median RT dose was 5040 (Range: 4500-5400) cGy and the median RT dose was 4500 (Range: 4500- 5040) cGy in those who did not (P = 0.014). The maximum dose of BM was 4927 cGy in patients with anemia and 4760 cGy in those who did not (P = 0.042). According to the classical statistical methods, age, RT, and BM maximum dose were found significant in terms of HT development.

Bone marrow V5, which is among the important variables in machine learning, was not statistically significant (P = 0.423).


  Discussion Top


Although IMRT increases dose conformity, it requires longer treatment time. VMAT has been shown to significantly improve treatment time.[18] Many studies have reported that VMAT provides a high-level conformal dose distribution, decreases doses to OAR, and shortens the duration of treatment compared to IMRT.[19] The long duration of treatment in IMRT contributes to patient discomfort during treatment and the increased monitor units (MUs) could lead to an increased incidence of radiation-related secondary cancers.[20],[21] According to the IMRT method (usually 5–7 fields are used), treatment with more fields (rotational therapy) in VMAT may be a cause that changes bone marrow toxicity.

In a study in which Cozzi et al. compared VMAT and 5-field IMRT in cervical cancer, similar target volume coverage was achieved in both plans, while better homogeneity and conformity were achieved with the VMAT technique. In addition, OAR (rectum, bladder, and small intestine) sparing was significantly improved with VMAT. This resulted in a potential relative reduction in normal tissue complication probability (NTCP) estimates for rectal bleeding, bladder contracture/volume loss, and small bowel obstruction/perforation by 30–70%. Compared to IMRT, the integral dose to healthy tissue was reduced by 12% with VMAT.[19] In a study, a total of 398 patients with stage IA–IVB cervical cancer treated with definitive RT with or without ChT were retrospectively analyzed (331 VMAT and 67 IMRT). A total prescription dose of 45–50 Gy was delivered to the pelvic field with VMAT/IMRT in 25/28 fractions, with five fractions per week. Every patient further received brachytherapy for four to six 6.0-Gy fractions. The median follow-up period was 25.47 (range: 0.93–58.93) months for the VMAT and 35.07 (range: 4.8–90.37) months for IMRT. The 3-year overall survival (OS), disease-free survival (DFS), local control (LC), and distant metastasis-free survival (DMFS) rate were 80.5, 65.4, 88.7, and 78.1% in the VMAT group, and 76.2, 76.4, 83.1, and 86.1% in the IMRT group, respectively. No significant differences were found between VMAT and IMRT groups for OS, DFS, LC, and DMFS rates. However, patients in the VMAT group had a lower incidence of chronic enterocolitis complications (26.6 versus 38.8%, P = 0.004).[22]

In the literature, previous studies suggest that BM volume exposed to lower doses can predict HT in patients undergoing pelvic RT.[10],[11] However, in the RTOG 0418 trial including 83 cases (43 endometrium and 40 cervical cancer), the patients were treated with postoperative IMRT of 50.4 Gy delivered to the pelvic lymphatics and vagina. The patients with cervical cancer weekly received a simultaneous cisplatin ChT. In the same study, 75% of patients with V40 >37% and 40% with V40 ≤37% had ≥grade 2 HT. The incidence of HT was statistically significant in patients with a median BM dose of >34.2 Gy (P = 0.049).[14] The cases in the RTOG 0418 study were treated with IMRT and the V40 values were significantly higher than the current study in which the VMAT technique was used and the median V40 was calculated as 5.5% (range: 0–22.5). Since the VMAT technique reduces the high dose BM volume, no statistically significant relationship was found with HT.

In a study evaluating cervical cancer patients, grade 3 HT was more common in those with BM V10 of ≥95% (68.8% versus 24.6%, P < 0.001).[11] In another study conducted with cervical cancer patients, a BM V20 of >80% was associated with ≥ grade 2 HT.[12]

The VMAT technique is known to increase the low-dose volume. In the present study, the V10-45 values were lower than those reported by IMRT and 3DCRT studies,[12],[22] but to the best of our knowledge, no studies in the literature have compared V5 or lower-dose volumes in gynecologic cancers.

BM is known to be affected by low radiation doses, such as 0.1 Gy.[23] However, given that RT is a local treatment, the expected level of BM toxicity may not occur in the doses encountered because of the continued contribution of the preserved BM areas to hematopoiesis. Therefore, it may not be the right approach to associate BM toxicity with dosimetric data alone. Clinical data should be taken into consideration in addition to dosimetric data when evaluating BM toxicity.

A better understanding of the dose-toxicity relationship is critical for a safer dose enhancement while improving local control in the radiotherapy of gynecological cancers. Assessment of toxicity due to RT with machine learning is a subject of interest recently. Machine learning and toxicity studies are available in prostate, lung, and head and neck cancers.[24],[25],[26] Valdes et al. used machine learning to investigate radiation pneumonitis (RP) in patients who underwent stereotactic body radiotherapy (SBRT) with stage I non-small cell lung cancer (NSCLC). A total of 201 patients were evaluated and RP was observed in 4%. Decision tree, Random Forest, and RUSBoost algorithms were used for machine learning. The authors stated that at least 800 cases should be included in the study in order to predict RP by machine learning with an error margin of 10% or less.[25] In a study by Dean et al., the patients who underwent RT due to head and neck cancer and developed severe acute oral mucositis were evaluated. Penalized Logistic Regression, Support Vector Classification, and Random Forest Classification algorithms were used and the best results were obtained by the Random Forest Classification algorithm and the most important factor affecting the severe acute mucositis was accepted as the dose received by the oral cavity.[26]

There is a limited number of studies evaluating RT-induced rectal toxicity by machine learning in cervical cancer.[27],[28] In the study by Zhen et al., 42 external RT along with brachytherapy cases were retrospectively evaluated, of whom 12 developed toxicity. It was reported that a rectal dose-toxicity estimation model could be established based on a Convolutional Neural Network model.[27]

In this study, it was aimed to predict hematological toxicity with machine learning in gynecological cancers applied adjuvant RT and age, KPS, metastatic lymph node presence, RT dose, ChT history, BM volume, BM V5, and BM minimum and maximum dose were determined as important variables. Parameters affecting hematologic toxicity were evaluated by also classical statistical methods and there was a statistically significant relationship between age, RT, and BM maximum dose.

To the best of our knowledge, there is no study in the literature investigating the hematological toxicity assessment of gynecologic cancers with machine learning techniques. Treatment-induced toxicity should be kept at an acceptable level while providing local control in patients. Although some recommended dose regimens have been provided for OAR, serious toxicities have been observed in some patients.[29] If cases with a high-risk of developing toxicity can be predicted, toxicity rates and gradations can be reduced.

Limitations of the study include contouring of OAR 1 cm above and below PTV and the low number of cases for assessment of machine learning. Machine learning's main strength comes from the sheer number of cases. The number of 121 cases is sufficient to begin with but insufficient to reach big data.

The main advantages of machine learning systems are: wide application area, high estimation ability, having clustering algorithms that reveal the basic structure in the dataset, classification algorithms that help determine the correct category, estimating possible errors and providing early intervention, generating time and resources, reducing cost, interpreting big data, and detection and reduction of unusual errors and risks. The major disadvantages of machine learning systems are big data requirements and overfitting.


  Conclusion Top


In gynecologic tumors treated with adjuvant RT with the VMAT technique, the factors that most affected acute HT were age, KPS, presence of metastatic lymph node, RT dose, ChT history, BM volume, BM V5, and BM minimum and maximum dose, and SVM was the best algorithm for side effect estimation. In the case of high treatment costs, potential serious toxicity and ineffective treatment, estimation with machine learning are interesting, given the damage of early progression and low survival. Studies with multicentered large data can provide algorithms with higher accuracy rates.

Trying to predict the future will benefit both the patient and the doctor.

Financial support and sponsorship

Nil.

Conflicts of interest

There are no conflicts of interest.



 
  References Top

1.
Mauch P, Constine L, Greenberger J, Knospe W, Sullivan J, Liesveld JL, et al. Hematopoietic stem cell compartment: Acute and late effects of radiation therapy and chemotherapy. Int J Radiat Oncol Biol Phys 1995;31:1319-39.  Back to cited text no. 1
    
2.
Alter BP. Diagnosis, Genetics, and Management of Inherited Bone Marrow Failure Syndromes. ASH Education Program Book 2007; 2007. p. 29–39.  Back to cited text no. 2
    
3.
Miura Y. Human bone marrow mesenchymal stromal/stem cells: Current clinical applications and potential for hematology. Int J Hematol 2016;103:122–8.  Back to cited text no. 3
    
4.
Cao X, Wu X, Frassica D, Yu B, Pang L, Xian L, et al. Irradiation induces bone injury by damaging bone marrow microenvironment for stem cells. Proc Natl Acad Sci USA 2011;108:1609-14.  Back to cited text no. 4
    
5.
Ellis RE. The distribution of active bone marrow in the adult. Phys Med Biol 1961;5:255–8.  Back to cited text no. 5
    
6.
Mell LK, Kochanski JD, Roeske JC, Haslam JJ, Mehta N, Yamada SD, et al. Dosimetric predictors of acute hematologic toxicity in cervical cancer patients treated with concurrent cisplatin and intensity-modulated pelvic radiotherapy. Int J Radiat Oncol Biol Phys 2006;66:1356–65.  Back to cited text no. 6
    
7.
Blomlie V, Rofstad EK, Skjonsberg A, Tvera K, Lien HH. Female pelvic bone marrow: Serial MR imaging before, during, and after radiation therapy. Radiology 1995;194:537–543.  Back to cited text no. 7
    
8.
Roeske JC, Lujan A, Rotmensch J, Waggoner SE, Yamada D, Mundt AJ. Intensity-modulated whole pelvic radiation therapy in patients with gynecologic malignancies. Int J Radiat Oncol Biol Phys 2000;48:1613-21.  Back to cited text no. 8
    
9.
Heron DE, Gerszten K, Selvaraj RN, King GC, Sonnik D, Gallion H, et al. Conventional 3D conformal versus intensity-modulated radiotherapy for the adjuvant treatment of gynecologic malignancies: A comparative dosimetric study of dose-volume histograms. Gynecol Oncol 2003;91:39-45.  Back to cited text no. 9
    
10.
Mell LK, Tiryaki H, Ahn KH, Mundt AJ, Roeske JC, Aydogan B. Dosimetric comparison of bone marrow-sparing intensity-modulated radiotherapy versus conventional techniques for treatment of cervical cancer. Int J Radiat Oncol Biol Phys 2008;71:1504-10.  Back to cited text no. 10
    
11.
Rose BS, Aydogan B, Liang Y, Yeginer M, Hasselle MD, Dandekar V, et al. Normal tissue complication probability modeling of acute hematologic toxicity in cervical cancer patients treated with chemoradiotherapy. Int J Radiat Oncol Biol Phys 2011;79:800-7.  Back to cited text no. 11
    
12.
Albuquerque K, Giangreco D, Morrison C, Siddiqui M, Sinacore J, Potkul R, et al. Radiation-related predictors of hematologic toxicity after concurrent chemoradiation for cervical cancer and implications for bone marrow-sparing pelvic IMRT. Int J Radiat Oncol Biol Phys 2011;79:1043-7.  Back to cited text no. 12
    
13.
Chen C, He M, Zhu Y, Shi L, Wang X. Five critical elements to ensure the precision medicine. Cancer Metastasis Rev 2015;34:313–8.  Back to cited text no. 13
    
14.
Kohane IS, Drazen JM, Campion EW. A glimpse of the next 100 years in medicine. N Engl J Med 2012;26:2538-9.  Back to cited text no. 14
    
15.
Lambin P, van Stiphout RGPM, Starmans MHW, Rios-Velazquez E, Nalbantov G, Aerts HJWL, et al. Predicting outcomes in radiation oncology– multifactorial decision support systems. Nat Rev Clin Oncol 2013;10:27–40.  Back to cited text no. 15
    
16.
Lim K, Small W Jr, Portelance L, Creutzberg C, Jürgenliemk-Schulz IM, Mundt A, et al. Consensus guidelines for delineation of clinical target volume for intensity- modulated pelvic radiotherapy for definitive treatment of cervix cancer. Int J Radiat Oncol Biol Phys 2011;79:348-55.  Back to cited text no. 16
    
17.
Powers D. Evaluation: From precision, recall and F-measure to ROC, informedness, markedness and correlation. J Mach Learn Technol 2011;2:2229-3981.  Back to cited text no. 17
    
18.
Otto K. Volumetric modulated arc therapy: IMRT in a single gantry arc. Med Phys 2008;35:3107.  Back to cited text no. 18
    
19.
Cozzi L, Dinshaw KA, Shrivastava SK, Mahantshetty U, Engineer R, Deshpande DD, et al. A treatment planning study comparing volumetric arc modulation with RapidArc and fixed field IMRT for cervix uteri radiotherapy. Radiother Oncol 2008;89:18091.  Back to cited text no. 19
    
20.
Dörr W, Herrmann T. Second primary tumors after radiotherapy for malignancies. Treatment-related parameters. Strahlenther Onkol 2002;178:357-62.  Back to cited text no. 20
    
21.
Hall EJ. Intensity-modulated radiation therapy, protons, and the risk of second cancers. Int J Radiat Oncol Biol Phys 2006;65:1-7.  Back to cited text no. 21
    
22.
Lin Y, Ouyang Y, Chen K, Lu Z, Liu Y, Cao X. Clinical outcomes of volumetric modulated arc therapy following intracavitary/interstitial brachytherapy in cervical cancer: A single institution retrospective experience. Front Oncol 2019;16:760.  Back to cited text no. 22
    
23.
Fujishiro A, Miura Y, Iwasa M, Fujii S, Sugino N, Andoh A, et al. Effects of acute exposure to low-dose radiation on the characteristics of human bone marrow mesenchymal stromal/stem cells. Inflamm Regen 2017;37:19.  Back to cited text no. 23
    
24.
Lee S, Kerns S, Ostrer H, Rosenstein B, Deasy JO, Oh JH. Machine learning on a genome-wide association study to predict late genitourinary toxicity after prostate radiation therapy. Int J Radiat Oncol Biol Phys 2018;101:128-35.  Back to cited text no. 24
    
25.
Valdes G, Solberg TD, Heskel M, Ungar L, Simone CB. Using machine learning to predict radiation pneumonitis in patients with stage I non-small cell lung cancer treated with stereotactic body radiation therapy. Phys Med Biol 2016;16:6105-20.  Back to cited text no. 25
    
26.
Dean JA, Wong KH, Welsh L, Jones AB, Schick U, Nwbold KL, et al. Normal tissue complication probability (NTCP) modelling using spatial dose metrics and machine learning methods for severe acute oral mucositis resulting from head and neck radiotherapy. Radiother Oncol 2016;120:21–7.  Back to cited text no. 26
    
27.
Zhen X, Chen J, Hrycushko B, Zhou L, Jiang S, Albuquerque K, et al. Deep convolutional neural network with transfer learning for rectum toxicity prediction in cervical cancer radiotherapy: A feasibility study. Phys Med Biol 2017;62:8246-63.  Back to cited text no. 27
    
28.
Chen J, Chen H, Zhong Z, Wang Z, Hrycushko B, Zhou L, et al. Investigating rectal toxicity associated dosimetric features with deformable accumulated rectal surface dose maps for cervical cancer radiotherapy. Radiat Oncol 2018;13:125.  Back to cited text no. 28
    
29.
Marks LB, Yorke ED, Jackson A, Ten Haken RK, Constine LS, Eisbruch A, et al. Use of normal tissue complication probability models in the clinic. Int J Radiat Oncol Biol Phys 2010;76:9-10.  Back to cited text no. 29
    


    Figures

  [Figure 1], [Figure 2], [Figure 3]
 
 
    Tables

  [Table 1], [Table 2], [Table 3], [Table 4], [Table 5], [Table 6]



 

Top
Previous Article  Next Article

    

  Site Map | What's new | Copyright and Disclaimer
  Online since 1st April '07
  2007 - Indian Journal of Cancer | Published by Wolters Kluwer - Medknow