Machine Learning-Based Model for Predicting Radiation Pneumonitis in Locally Advanced Non- Small Cell Lung Cancer Treated with IMRT-A Two-Centre Study

doi:10.21203/rs.3.rs-4920517/v1

Download PDF

Research Article

Machine Learning-Based Model for Predicting Radiation Pneumonitis in Locally Advanced Non- Small Cell Lung Cancer Treated with IMRT-A Two-Centre Study

https://doi.org/10.21203/rs.3.rs-4920517/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Aim: To build and externally validate machine learning-based models for radiation pneumonitis (RP) prediction in patients with locally advanced non-small cell lung cancer (LA-NSCLC) treated with intensity-modulated radiation therapy (IMRT) in the era of precision radiotherapy.

Patients and Methods: In this two-center retrospective study, a total of 218 patients (131 in the training cohort, and 87 in the external validation cohort) with LA-NSCLC. All patients underwent primary IMRT with strict lung dose constraints. Pretreatment CT radiomics features were extracted and then generated radiomics score (Rad-score). The study factors included Rad-score, dose-volume parameters and clinical features. Based on the independent risk factors, three machine learning models (random forest, logistic regression and decision tree) were developed and validated for predicting RP. The predictive performances of the models were evaluated using area under receiver operating characteristic curve (AUC) and decision curve analysis (DCA).

Results: Within both cohorts, the overwhelming majority of patients were safely treated with radiotherapy within known lungs dose constraints. PE,ILD,N2-N3, ipsilateral lung Rad-score and contralateral lung Rad-score were independent risk factors for RP (P＜0.05). The AUC of random forest model, logistic regression model and decision tree model were 0.938, 0.859 and 0.632 in the training cohort, and 0.885, 0.911 and 0.721 in the external validation cohort, respectively. The calibration curve and DCA demonstrated goodness-of-ft and improved benefits in random forest model.

Conclusion: PE, ILD, N2-N3 and CT radiomics features of lungs were independent predictors of RP in the LA-NSCLC patients treated with IMRT. The model using random forest algorithm exhibited the best predictive accuracy, outperforming logistic regression and decision tree.

non-small cell lung cancer

machine learning

radiation pneumonia

radiomics

Non-small cell lung cancer (NSCLC) accounts for 85% of all lung malignancies and carries a poor prognosis^{[1, 2]}. More than 50% of the patients diagnosed with NSCLC present locally advanced disease (LA-NSCLC) at onset, which is often unresectable. Concurrent chemoradiotherapy (CCRT) is the standard treatment for these patients^[3–5]. Radiation pneumonia (RP) is the most common dose-limiting complication in patients with lung cancer receiving thoracic radiotherapy(RT)^[6]. The incidence of symptomatic RP has been reported to be about 20–30% in NSCLC patients treated with CCRT^{[7, 8]}. RP has been shown to be a serious burden for patients and can even be life-threatening.

In a pre-intensity-modulated radiation therapy (IMRT) era, considerable efforts have been made to establish predictive variables and guide prevention efforts to reduce the incidence of RP. In particular, dosimetric parameters were confirmed to be risk factors for the development of RP, including mean lung dose and dose volume histogram parameters such as the percent volume of the total lung receiving more than 5 Gy, the percent volume of the total lung receiving more than 20 Gy and the percent volume of the total lung receiving more than 30 Gy (named V5, V20 and V30) ^[9–15]. Individual factors such as age, chronic obstructive pulmonary disease (COPD) and interstitial lung disease had also been shown to be risk factors for the development of RP^{[16, 17]}. With the development of radiotherapy technology, IMRT has become the first radiotherapy choice for LA-NSCLC because of its accuracy and safety. In situations when 3D conformal planning is unable to meet dosimetric constraints for the lungs, IMRT can help meet the dose constraints for these structures with its dosimetric advantages in radiotherapy planning, which reduce the incidence of RP^{[18, 19]}. However, grade ≥ 2 RP occurred irrespective of the application of dose constraints and the use of sophisticated delivery technique. This suggests that previous findings on RP risk factors may be less generalizable in the era of individualized precision radiotherapy and highlights the need to further explore novel predictors of RP risk.

Radiomics is the extraction of quantitative imaging features from medical images, which has the important advantage of being nonintrusive and highly specific. Radiomics has been proposed to explore the correlation between medical images and underlying individual characteristics, to improve predictive accuracy and guide clinical decisions^[20]. The rapid development and validation of radiomics signatures for precision diagnosis and treatment provide a powerful tool in modern personalized medicine^[21]. A radiomic approach has been proven to be a critical supplement to the clinical parameters and dose-volume indices in predicting RP^{[22, 23]}. As a branch of artificial intelligence, machine learning technology is helpful to establish a robust model and has been introduced as research tools for cancer diagnosis, prognosis and treatment selection^[24]. Recently, an increasing number of researches have applied the machine learning into the radiotherapy fields, such as machine learning radiotherapy dose prediction^[25–27] and automatic treatment planning^[28–30]. Nonetheless, prediction models of RP using these novel approaches remains relatively limited, warranting further exploration.

In the current study, we aim to construct models based on CT radiomics features by using machine learning for predicting the incidence of RP following IMRT under strict lung dose constraints.

Patients

The flow chart is presented in Fig. 1. All patients were initially diagnosed with LA-NSCLC according to the AJCC 8th edition. Inclusion criteria were as follows: (1) underwent primary IMRT with strict lung dose constraints; (2) RT dose prescription ≥ 56 Gy and completed the RT; (3) estimated survival greater than 6 months after RT; (4) Complete follow-up information was available. Exclusion criteria were as follows:(1) low image quality; (2) previous history of thoracic RT and surgery.

All patients received definitive concurrent or sequential CRT. IMRT was employed. Gross tumor volume (GTV) was contoured in sequential axial CT images and included the primary tumor and metastatic lymph node drainage area. The clinical target volume (CTV) was created by expanding the GTV by 0.6–0.8cm. Dosimetrist were instructed to cover at least 95% of the planning target volume (PTV) with the prescribed RT dose. Delineations of GTV, CTV, PTV and organ-at‐risks (OARs) including lung, heart, spinal cord and esophagus conformed to the guidelines set by the Radiotherapy and Oncology Group lung cancer delineation guidelines (RTOG 1106). The median total dose was 60 Gy (range, 56–64 Gy). The fraction dose was 1.8–2.25 Gy (median: 2.0 Gy). The dose constraints of OARs were as follows: maximum point dose of spinal cord < 45 Gy; total lungs V5 < 60%, V20 < 30%, V30 < 20%, mean lung dose (MLD) < 20 Gy; heart V30 < 40%, V40 < 30%.

Clinical and dosimetric data

Clinical data, including characteristics and dosimetric factors, were collected and analyzed. General clinical characteristics of patients: gender, age, smoking status, clinical stages, PE and ILD. The dosimetric parameters were extracted from the treatment planning system, including the radiation dose, PTV volume, MLD, V5, V10, V20, V30 and V40.

CT image acquisition and radiomics features extraction

Intravenous contrast-enhanced planning CT scans were acquired on a large-aperture spiral CT scanner (Light SpeedRT, GE) multislice scanner with a standardized protocol (140 kV, 380 mA, 2.5mm slice thickness, 512 *512 image matrix). The scanning region was from the skull base to the 4th lumbar level. The CT scan images were then transferred to the MIM planning system (MIM Software Inc.) in DICOM format. Images with substantial CT artifacts were excluded to avoid confounding on sub-sequent feature extraction and analysis. The regions of interest (ROIs) were manually outlined layer by layer on CT images by a radiologist who had 10 years of experience using the MIM planning system. After the first seg-mentation, a second radiologist who had more than 20 years of experience checked the accuracy. All differences were resolved by discussion.

We measured radiomic features from the ipsilateral lung of the tumor (ipsilateral-lung) and the contralateral lung of the tumor (contralateral-lung). All features were extracted by using the radiomics plug-in in the 3D Slicer (version 5.2.1; http://www.slicer.org), a free open-source software application for medical image computing. Each set of images extracted a total of 1037 features, including first-order features, shape features, texture features, gray level cooccurrence matrix (GLCM), gray level dependence matrix (GLDM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM) and neighboring gray tone difference matrix (NGTDM), based on both the original and derived images. Then, the least absolute shrinkage and selection operator (LASSO) was used to select the optimal radiomics features subset to construct the radiomics score (Rad-score).

Evaluation of RP

Patients were followed up at 1 month after radiotherapy and then every 2–3 months thereafter. Physical examination, blood tests and chest CT scans were performed at each follow-up. The follow-up time was shorter if pneumonitis was suspected when patients developed dry cough, dyspnea, or fever. RP was diagnosed by clinical symptoms and the characteristic imaging findings on CT. Grading of RP was conducted according to the Common Terminology Criteria for Adverse Events version 5.0 (CTCAE v5.0). The diagnosis and grading of RP were confirmed by two experienced radiation oncologists.

Machine learning modeling and statistical analyses.

Continuous variables were expressed as median ± interquartile range, and categorical variables were expressed as numbers and percentages. We employed univariate analysis and multivariate logistic regression analysis to access the independent effect of clinical and dosimetric features in the prediction of RP. Logistic regression, decision tree and random forest classifier were based on the "rms", "rpart" and "randomForest" R language package, respectively. These models were subsequently validated in an external dataset. Corresponding AUCs with 95% CI, sensitivity, and specificity were calculated. And DCA was applied to assess the clinical practical value. All statistical analyses were based on R 4.2.3. A two-sided p < 0.05 defined statistical significance.

Patient characteristics and incidence of RP

A total of 131 patients from the Guangxi Medical University Cancer Hospital treated between July 2020 and June 2023 were included in the training cohort and an additional 87 patients from the Chongqing University Cancer Hospital treated between July 2016 and June 2020 were included in the external validation cohort (Table 1). Within both cohorts, the vast majority of patients were safely treated with radiotherapy within the predefined stringent lungs dose constraints. In particular, no violations at V20 and MLD thresholds were observed in the included patients, with a mean V20 of 25.4% and a mean MLD of 13.94 Gy.

Table 1

Basic information of the patients in training cohort and external validation cohorts.
Characteristic	Training cohort (n = 131)	External validation cohort (n = 87)
RP
RP < 2	107(81.7%)	76(87.4%)
RP ≥ 2	24(18.3%)	11(12.6%)
Days from onset of RT to RP(d)	60(28–82)	65 (42–75)
Age
<60	51(38.9%)	34(39.1%)
≥ 60	80(61.0%)	53(60.9%)
Gender
Male	97(74.0%)	44(50.6%)
Female	34(26.0%)	43(49.4%)
Smoking
Current or Former	78(59.5%)	59(67.8%)
Never	53(40.5%)	28(32.2%)
T Stage
1–2	43(32.8%)	30(34.5%)
3–4	88(67.2%)	57(65.5%)
N Stage
0–1	21(16.0%)	10(11.5%)
2–3	110(84.0%)	77(88.5%)
PE
≤ 1	101(77.1%)	66(75.9%)
>1	30(%22.9)	21(24.1%)
ILD
0	121(92.4%)	81(93.1%)
≥ 1	10(7.6%)	6(6.9%)
Dosimetric parameters
PTV (cm³)	449.78(240.56–636.80)	298.90(199.45-404.46)
Fraction dose (Gy)	2.00(2.00–2.00)	2.00(2.00–2.00)
Total radiation dose (Gy)	60.00(60.00–60.00)	60.00(50.00–60.00)
Lung V5 (%)	50.6(42.59–57.49)	47.92(38.39–55.75)
Lung V10 (%)	37.56(29.72–41.49)	33.26(27.03–41.18)
Lung V20 (%)	25.41(20.53–27.56)	20.82(16.30-25.09)
Lung V30 (%)	18.24(12.85–19.72)	13.58(10.08–17.37)
Lung V40 (%)	11.86(7.84–13.94)	8.18(6.03–11.76)
MLD (Gy)	13.94(10.68–15.13)	11.66(9.16–13.95)

In the training and validation group, the average follow-up time of CT was 8 and 9 months, and RP occurred in 24 (18.3%) and 11 (12.6%) patients, respectively. In the training group, there were 17 patients (13.0%), 5 patients (3.8%) and 2 patient (1.5%) with grade 2, 3 and 4 RP, respectively. The median time from completion of radiotherapy to the occurrence of ≥ grade 2 RP (symptomatic RP) was 60 days in the training group and 65 days in the validation group.

Radiomics features selection

Each of 1037 radiomic features were extracted from the selected volume of interest (VOI) of the ipsilateral lung and the contralateral lung. Four radiomic features of the ipsilateral lung (LLH-firstorder-Minimum, HLL-firstorder-Skewness, HLL-glcm-MCC and HHH-firstorder-Median), and two radiomic features of the contralateral lung (HHH-glszm-LargeAreaEmphasis and HHH-glszm-ZoneVariance) were screened out using LASSO model (Fig. 2) and Rad-scores were built separately, where the coefficients were utilized for weighting (refer to Fig. 3).

Development and validation of predictive models

The results of the univariate analysis and multivariate LR analysis in the training cohort are shown in Table 2. Spearman correlation analysis showed no significant correlation among The N(N2-N3) Stage, PE, ILD, ipsilateral Radscore and contralateral Radscore (Fig. 4). Subsequently, we established random forest model, logistic regression model (Fig. 5) and decision tree model, random forest and decision tree selected the optimal parameters through cross-validation. And then conducted external validation.

Table 2

Univariate analysis and multivariate analysis for risk factors of grade ≥ 2 RP.
Features	Univariate analysis		Multivariate analysis
Features	OR (95%CI)	P value	OR (95% CI)	P value
Age	2.86(0.99–8.24)	0.051
Gender	0.51(0.16–1.63)	0.257
Gender	0.51(0.16–1.63)	0.257
Smoking	1.45(0.57–3.68)	0.433
T Stage	1.59(0.57–3.68)	0.369
N Stage	0.21(0.07–0.58)	0.003*	0.03(0.003–0.27)	0.002*
PE	4.94(1.91–12.75)	<0.001*	12.01(2.49–57.99)	0.002*
ILD	14.27(3.36–60.64)	<0.001*	38.44(3.27-451.01)	0.004*
PTV	1.00(0.99-1.00)	0.722
Fraction dose	1.25(0.60–1.61)	0.536
Total radiation dose	o.99(0.84–1.16)	0.898
Lung V5	0.98(0.95–1.02)	0.284
Lung V10	0.96(0.92–1.01)	0.133
Lung V20	0.95(0.89–1.03)	0.253
Lung V30	0.97(0.89–1.06)	0.535
Lung V40	0.99(0.89–1.11)	0.904
MLD	0.99(0.99-1.00)	0.328

The AUCs of the training cohort for random forest model, logistic regression model and decision tree model were 0.938 (95% confidence interval (CI): 0.898–0.983), 0.859 (95%CI: 0.774–0.931) and 0.632 (95%CI: 0.537–0.726), respectively. The AUC of the external validation cohort for these three models were 0.885 (95%CI: 0.772–0.999), 0.911 (95%CI: 0.878–0.941) and 0.721 (95%CI: 0.565–0.875), respectively. Figure 6 shows the ROC curves of the three models and Table 3 lists the AUC, accuracy, sensitivity, specificity, PPV, and NPV of each model in the training, and external validation cohorts. The random forest model was identified as the most effective model. Favorable calibrations of random forest model were demonstrated in the training, and validation cohorts (Fig. 7). The DCA revealed that using random forest model to predict RP can gain more benefit than logistic regression model and decision tree model when the threshold probability is from 0–66% in the training cohort, and 33–56% in the external validation cohort (Fig. 8).

Table 3

Evaluation and comparison of the predictive ability of models.
Model	AUC (95% CI)	Accuracy	Sensitivity	Specificity	PPV	NPV
Training cohort
Random forest	0.938 (0.898–0.983)	0.735	0.972	0.500	0.897	0.800
Logistic regression	0.859 (0.774–0. 931)	0.683	0.990	0.375	0.876	0.900
Decision tree	0.632 (0.537–0.726)	0.632	0.971	0.291	0.859	0.700
Validation cohort
Random forest	0.885 (0.772–0.999)	0.707	0.960	0.455	0.924	0.625
Logistic regression	0.911 (0.878–0.941)	0.500	1.00	0.00	0.873	-
Decision tree	0.721 (0. 565- 0. 875)	0.720	0.986	0.454	0.925	0.833

Based on our results, N Stage, PE, ILD, ipsilateral Radscore and contralateral Radscore were independent risk factors for the development of RP in LA-NSCLC patients after IMRT. Using three machine learning algorithms, we built three models based on these features to predict RP, and selected the random forest model as the most optimal model according to training and external validation results.

Dose-volume indices have been studied for decades and they are widely known as significant risk factors for RP. In particular, lung mean and lung V20 are generally recognized criteria^[31]. However, we did not observe an association between the lung dosimetric parameter and the risk of RP in our cohort, which seemed to be against consensus. Several similar studies in recent years reached the same conclusion^{[32, 33]}. The results could be explained by the technological advancement, which made the influence of dosimetric parameters minimized to a non-significant level. IMRT is a more sophisticated techniques of conformal RT technologies and is associated with a lower incidence of RILI compared with standard 3D-CRT. Based on previous research results from 3D-CRT era^[10], strict lung dose-volume constraint is now particularly emphasized in the IMRT era. Previous studies ^[34]suggested that the V5 of total lung should be maintained at < 60–65%, V20 should be maintained at < 30% and mean lung dose < 20 Gy to reduce the risk of RP^[9–15]. Within both cohorts in our study, no violations at V20 and MLD thresholds and only a few and mild surpasses at V5 and V30 were observed in the included patients. These results reflect to some extent our ability to implement precision radiotherapy by using IMRT technique. In this case, the conventional dosimetric parameters may no longer be useful as independent predictors for predicting RP. Although the variations of the above-mentioned dose-volume indices among individuals are minimized by the application of strict dose–volume constraints, the spatial dose distributions can vary widely due to different radiotherapy plans. Several studies had been conducted to explore more deeply the interrelated effects of dose and morphology. It is reported recently that spatial dose distributions, instead of conventional dosimetric parameters, are intimately correlate with the incidence RP^{[22, 35, 36]}. On the other hand, individual variations warrant more concern in the estimation of RP risk when dose-volume indices are severely restricted to prevent RP.

In the current study, the N stage, ILD grade, PE grade had been confirmed to be closely related to the occurrence of RP ^{[14, 17, 37]}. Okubo M suggested a significant association between subclinical ILD and the occurrence of grade 2 or higher radiation pneumonitis^[38]. Zhou and team observed a strong correlation between the severity of PE and the higher cumulative incidence of symptomatic or severe radiation pneumonitis in locally advanced NSCLC^[39]. Thoracic radiation may induce lymphocyte-mediated hypersensitivity reactions, which could exacerbate acute ILD. Therefore, accurate pre-radiotherapy localization of the treatment area and precise assessment of ILD and PE are crucial. In patients with stage N2 and N3 NSCLC, the higher tumor burden often indicates a more complex pathophysiological process. Definitive radiotherapy must cover a wide area to ensure adequate treatment of both the tumor and its lymphatic metastases^[40]. The extensive mediastinal lymph node metastasis (N2-N3) leads to higher doses of radiation being delivered to portions of the lung tissue. However, this unavoidably increases the risk of irradiating part of normal lung tissue. The CT radiomics signature of lungs may represent different lung phenotype resulting from underlying lung diseases, the baseline lung status and the underlying radiosensitivity of lung tissue^{[41, 42]}. Previous studies have reported that CT-based radiomics features can help to predict pulmonary toxicity for NSCLC patients receiving radiotherapy or immune checkpoint inhibitor ^{[22, 36, 43]}. In this study, radiomics method was applied to extract large numbers of radiomics features from pretreatment CT images. The addition of CT radiomics signature provided a more comprehensive measurement of individual differences and improved the ability of models to predict RP with higher AUC values.

In this study, we built three models through different machine learning algorithms. The random forest model yielded excellent AUC values of 0.938 for the training set and 0.885 for the external validation set, outperformed the traditional logistic regression algorithm and decision tree algorithm. Machine learning, a scientific discipline that enables computers to learn from experiences, has recently been shown to have better performances over traditional statistical modeling approaches^[44–46]. The random forest is a machine learning method based on an ensemble of decision trees generated by random feature sets. It has been introduced as an accurate machine learning method the prediction of radiation-induced toxicities ^{[47, 48]}. In a study of 203 patients with LA-NSCLC, performed multivariate analysis using random forest to assess the combined performance of important predictors of RP with an AUC of 0.66 (p = 0.0005) ^[47]. According to a recent secondary analysis of RTOG 0617 to identify dosimetric predictors of toxicity in patients with LA-NSCLC using machine learning and artificial intelligence, the random forest model showed the best ability to predict the pulmonary toxicities with an AUC of 0.706, outperforming logistic regression^[48]. Our random forest model seems to have performed better than the previous established models using the same methods and other machine learning approaches ^{[35, 49]}.

Our study has several limitations. First, this was a small sample retrospective study, a selection bias may exist. However, participant recruitment was conducted at two institutions in this study, which may increase some representativeness. It is necessary to validate these models in larger datasets to stabilize our models. Second, even though random forest model could be used to predict multiple outcomes, our model can only predict the presence or absence of RP ≥ 2 but not each grade. This is because the low diversity of RP grades does not allow sufficient statistical power to classify RP grades. Third, only risk factors at the clinical level were included in the established model. In fact, the mechanism underlying RP is multifaceted and needs deeper and more extensive research. Recently, predicting models integrating other omics such as proteomics and genomics have been established^{[32, 50, 51]}. Integration of multi-omics and machine learning is an important next step for accurate RP risk prediction.

The N stage, ILD grade, PE grade and CT radiomics features of lungs were independent predictors of RP in the LA-NSCLC patients treated with IMRT. The model using the random forest algorithm exhibited the best predictive accuracy, outperforming logistic regression algorithm and decision tree algorithm.

RP, radiation pneumonitis; LA-NSCLC, locally advanced non-small cell lung cancer; Rad-score, radiomics score; CCRT, Concurrent chemoradiotherapy; IMRT, intensity-modulated radiation therapy; COPD, chronic obstructive pulmonary disease; ILD, interstitial lung disease; PE, pulmonary emphysema; 3D-CRT, 3 Dimensional Conformal Radio-Therapy; GTV, Gross tumor volume; CTV, clinical target volume; PTV, planning target volume; OARs, organ‐at‐risks; MLD, mean lung dose; ROIs, regions of interest; LASSO, the least absolute shrinkage and selection operator; ROC, Receiver Operation Characteristics; AUC, area under the curve.

Ethics approval and patient consent: The study has been approved by the Guangxi Medical University Cancer Hospital Ethical Review Committee (Approval Number: LW2024016), and complies with relevant ethical standards and regulations. Patient consent was not required because of the retrospective nature of the study.

Conflicts of Interest

The authors declare no conflict of interest.

Authors’ Contributions

Liu F and Yang D were responsible for data curation, methodology, visualization and writing e original draft preparation. Li L was responsible for software, data curation and methodology. Su T was responsible for data curation and software. Wu Q was responsible for data collection. Liang S was responsible for conceptualization, writing e reviewing and editing.

Acknowledgements

Not applicable.

Funding

This research was supported by the National Natural Science Foundation of China (82060576) and the Guangxi Key Research and Development Project (AB20159024).

Data Availability Statement

The statistical codes and datasets used and/or analyzed during the current study are available from the corresponding author ([email protected]) on reasonable request.

Siegel RL, Miller KD, Wagle NS, et al. Cancer statistics, 2023. Ca-a Cancer. J Clin. 2023;73:17–48. http://doi.org/10.3322/caac.21763.
Lee NSY, Shafiq J, Field M, et al. Predicting 2-year survival in stage I-III non-small cell lung cancer: the development and validation of a scoring system from an Australian cohort. Radiat Oncol. 2022;17:12. http://doi.org/10.1186/s13014-022-02050-1.
Perez CA, Stanley K, Rubin P, et al. A prospective randomized study of various irradiation doses and fractionation schedules in the treatment of inoperable non-oat-cell carcinoma of the lung. Preliminary report by the Radiation Therapy Oncology Group. Cancer. 1980;45:2744–53. .http://doi.org/10.1002/1097-0142(19800601)45:11<2744::Aid-cncr2820451108>3.0.Co;2-u.
Ettinger DS, Wood DE, Aisner DL, et al. Non-Small Cell Lung Cancer, Version 2.2021 Featured Updates to the NCCN Guidelines. J Natl Compr Canc Netw. 2021;19:254–66. http://doi.org/10.6004/jnccn.2021.0013.
Rossi G, Nappi O. What's new in the WHO classification of tumors of lung and pleura. Pathologica 110:3–4, 2018.http://doi.org/.
Gaspar LE. RADIATION PULMONARY TOXICITY: PREDICTION AND PREVENTION. J Thorac Oncol. 2011;6:S214–214. http://doi.org/.
Giridhar P, Mallick S, Rath GK et al. Radiation induced lung injury: prediction, assessment and management. Asian Pacific journal of cancer prevention: APJCP 16:2613-7, 2015.http://doi.org/.
Wang JB, Zhou ZM, Liang J, et al. Intensity-Modulated Radiation Therapy May Improve Local-Regional Tumor Control for Locally Advanced Non-Small Cell Lung Cancer Compared With Three-Dimensional Conformal Radiation Therapy. Oncologist. 2016;21:1530–7. http://doi.org/10.1634/theoncologist.2016-0155.
Song CH, Pyo H, Moon SH, et al. Treatment-related pneumonitis and acute esophagitis in non-small-cell lung cancer patients treated with chemotherapy and helical tomotherapy. Int J Radiat Oncol Biol Phys. 2010;78:651–8. .http://doi.org/10.1016/j.ijrobp.2009.08.068.
Hanania AN, Mainwaring W, Ghebre YT, et al. Radiation-Induced Lung Injury: Assessment and Management. Chest. 2019;156:150–62. http://doi.org/10.1016/j.chest.2019.03.033.
Rodrigues G, Lock M, D'Souza D, et al. Prediction of radiation pneumonitis by dose - volume histogram parameters in lung cancer–a systematic review. Radiother Oncol. 2004;71:127–38. http://doi.org/10.1016/j.radonc.2004.02.015.
Kocak Z, Borst GR, Zeng J, et al. Prospective assessment of dosimetric/physiologic-based models for predicting radiation pneumonitis. Int J Radiat Oncol Biol Phys. 2007;67:178–86. http://doi.org/10.1016/j.ijrobp.2006.09.031.
Ramella S, Trodella L, Mineo TC, et al. Adding ipsilateral V20 and V30 to conventional dosimetric constraints predicts radiation pneumonitis in stage IIIA-B NSCLC treated with combined-modality therapy. Int J Radiat Oncol Biol Phys. 2010;76:110–5. .http://doi.org/10.1016/j.ijrobp.2009.01.036.
Palma DA, Senan S, Tsujino K, et al. Predicting radiation pneumonitis after chemoradiation therapy for lung cancer: an international individual patient data meta-analysis. Int J Radiat Oncol Biol Phys. 2013;85:444–50. http://doi.org/10.1016/j.ijrobp.2012.04.043.
Bledsoe TJ, Nath SK, Decker RH. Radiation Pneumonitis. Clin Chest Med. 2017;38:201–8. .http://doi.org/10.1016/j.ccm.2016.12.004.
Takeda A, Kunieda E, Ohashi T, et al. Severe COPD is correlated with mild radiation pneumonitis following stereotactic body radiotherapy. Chest. 2012;141:858–66. http://doi.org/10.1378/chest.11-1193.
Ueki N, Matsuo Y, Togashi Y, et al. Impact of pretreatment interstitial lung disease on radiation pneumonitis and survival after stereotactic body radiation therapy for lung cancer. J Thorac Oncol. 2015;10:116–25. http://doi.org/10.1097/JTO.0000000000000359.
Bajcsay A, Janvary LZ, Ladanyi K et al. Application of modern radiotherapy in lung cancer. Magyar onkologia 64:255–261, 2020.http://doi.org/.
Guo TT, Zou LQ, Ni JJ, et al. Radiotherapy for unresectable locally advanced non-small cell lung cancer: a narrative review of the current landscape and future prospects in the era of immunotherapy. Translational Lung Cancer Res. 2020;9:2097–112. http://doi.org/10.21037/tlcr-20-511.
Lambin P, Rios-Velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012;48:441–6. http://doi.org/10.1016/j.ejca.2011.11.036.
Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol. 2017;14:749–62. http://doi.org/10.1038/nrclinonc.2017.141.
Zhang Z, Wang Z, Yan M, et al. Radiomics and Dosiomics Signature From Whole Lung Predicts Radiation Pneumonitis: A Model Development Study With Prospective External Validation and Decision-curve Analysis. Int J Radiat Oncol Biol Phys. 2023;115:746–58. http://doi.org/10.1016/j.ijrobp.2022.08.047.
Jiang W, Song Y, Sun Z, et al. Dosimetric Factors and Radiomics Features Within Different Regions of Interest in Planning CT Images for Improving the Prediction of Radiation Pneumonitis. Int J Radiat Oncol Biol Phys. 2021;110:1161–70. http://doi.org/10.1016/j.ijrobp.2021.01.049.
Tran KA, Kondrashova O, Bradley A, et al. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med. 2021;13:152. http://doi.org/10.1186/s13073-021-00968-x.
Bai X, Liu Z, Zhang J, et al. Comparing of two dimensional and three dimensional fully convolutional networks for radiotherapy dose prediction in left-sided breast cancer. Sci Prog. 2021;104:368504211038162. http://doi.org/10.1177/00368504211038162.
Chen X, Men K, Li Y, et al. A feasibility study on an automated method to generate patient-specific dose distributions for radiotherapy using deep learning. Med Phys. 2019;46:56–64. http://doi.org/10.1002/mp.13262.
Nguyen D, Jia X, Sher D, et al. 3D radiotherapy dose prediction on head and neck cancer patients with a hierarchically densely connected U-net deep learning architecture. Phys Med Biol. 2019;64:065020. http://doi.org/10.1088/1361-6560/ab039b.
Wang H, Bai X, Wang Y, et al. An integrated solution of deep reinforcement learning for automatic IMRT treatment planning in non-small-cell lung cancer. Front Oncol. 2023;13:1124458. http://doi.org/10.3389/fonc.2023.1124458.
Yang Y, Shao K, Zhang J, et al. Automatic Planning for Nasopharyngeal Carcinoma Based on Progressive Optimization in RayStation Treatment Planning System. Technol Cancer Res Treat. 2020;19:1533033820915710. http://doi.org/10.1177/1533033820915710.
Bai X, Shan G, Chen M, et al. Approach and assessment of automated stereotactic radiotherapy planning for early stage non-small-cell lung cancer. Biomed Eng Online. 2019;18:101. http://doi.org/10.1186/s12938-019-0721-7.
LBA35 Camrelizumb(C) plus rivoceranib (R) vs. sorafenib (S) as first-line therapy for unresectable hepatocellular carcinoma (uHCC): a randomized, phase III trial. ESMO. 2022.http://doi.org/.
Du L, Ma N, Dai X, et al. Precise prediction of the radiation pneumonitis in lung cancer: an explorative preliminary mathematical model using genotype information. J Cancer. 2020;11:2329–38. http://doi.org/10.7150/jca.37708.
Krafft SP, Rao A, Stingo F, et al. The utility of quantitative CT radiomics features for improved prediction of radiation pneumonitis. Med Phys. 2018;45:5317–24. http://doi.org/10.1002/mp.13150.
Liu Z, Liu W, Ji K, et al. Simultaneous integrated dose reduction intensity-modulated radiotherapy applied to an elective nodal area of limited-stage small-cell lung cancer. Exp Ther Med. 2015;10:2083–7. http://doi.org/10.3892/etm.2015.2835.
Zhang Z, Wang Z, Luo T, et al. Computed tomography and radiation dose images-based deep-learning model for predicting radiation pneumonitis in lung cancer patients after radiation therapy. Radiother Oncol. 2023;182:109581. http://doi.org/10.1016/j.radonc.2023.109581.
Kraus KM, Oreshko M, Bernhardt D, et al. Dosiomics and radiomics to predict pneumonitis after thoracic stereotactic body radiotherapy and immune checkpoint inhibition. Front Oncol. 2023;13:1124592. http://doi.org/10.3389/fonc.2023.1124592.
Zhou Z, Song X, Wu A, et al. Pulmonary emphysema is a risk factor for radiation pneumonitis in NSCLC patients with squamous cell carcinoma after thoracic radiation therapy. Sci Rep. 2017;7:2748. http://doi.org/10.1038/s41598-017-02739-4.
Okubo M, Itonaga T, Saito T, et al. Predicting risk factors for radiation pneumonitis after stereotactic body radiation therapy for primary or metastatic lung tumours. Br J Radiol. 2017;90. http://doi.org/10.1259/bjr.20160508.
Wu AL, Zhou ZY, Song YP, et al. Application of a radiation pneumonitis prediction model in patients with locally advanced lung squamous cell cancer. Annals Palliat Med. 2021;10:4409–17. http://doi.org/10.21037/apm-21-459.
Tahara M, Fuse N, Mizusawa J, et al. Phase I/II trial of chemoradiotherapy with concurrent S-1 and cisplatin for clinical stage II/III esophageal carcinoma (JCOG 0604). Cancer Sci. 2015;106:1414–20. http://doi.org/10.1111/cas.12764.
Xu Y, van Beek EJ, Hwanjo Y, et al. Computer-aided classification of interstitial lung diseases via MDCT: 3D adaptive multiple feature method (3D AMFM). Acad Radiol. 2006;13:969–78. http://doi.org/10.1016/j.acra.2006.04.017.
Hoffman EA, Reinhardt JM, Sonka M, et al. Characterization of the interstitial lung diseases via density-based and texture-based analysis of computed tomography images of lung structure and function. Acad Radiol. 2003;10:1104–18. .http://doi.org/10.1016/s1076-6332(03)00330-1.
Qiu Q, Xing L, Wang Y et al. Development and Validation of a Radiomics Nomogram Using Computed Tomography for Differentiating Immune Checkpoint Inhibitor-Related Pneumonitis From Radiation Pneumonitis for Patients With Non-Small Cell Lung Cancer. Front Immunol 13:8708422022.http://doi.org/10.3389/fimmu.2022.870842
Beam AL, Kohane IS. Big Data and Machine Learning in Health Care. JAMA. 2018;319:1317–8. http://doi.org/10.1001/jama.2017.18391.
Chen JH, Asch SM. Machine Learning and Prediction in Medicine - Beyond the Peak of Inflated Expectations. N Engl J Med. 2017;376:2507–9. http://doi.org/10.1056/NEJMp1702071.
Goldstein BA, Navar AM, Carter RE. Moving beyond regression techniques in cardiovascular risk prediction: applying machine learning to address analytic challenges. Eur Heart J. 2017;38:1805–14. http://doi.org/10.1093/eurheartj/ehw302.
Luna JM, Chao HH, Diffenderfer ES, et al. Predicting radiation pneumonitis in locally advanced stage II-III non-small cell lung cancer using machine learning. Radiother Oncol. 2019;133:106–12. http://doi.org/10.1016/j.radonc.2019.01.003.
Ladbury C, Li R, Danesharasteh A, et al. Explainable Artificial Intelligence to Identify Dosimetric Predictors of Toxicity in Patients with Locally Advanced Non-Small Cell Lung Cancer: A Secondary Analysis of RTOG 0617. Int J Radiat Oncol Biol Phys. 2023. http://doi.org/10.1016/j.ijrobp.2023.06.019.
Yakar M, Etiz D, Metintas M, et al. Prediction of Radiation Pneumonitis With Machine Learning in Stage III Lung Cancer: A Pilot Study. Technol Cancer Res Treat. 2021;20:15330338211016373. http://doi.org/10.1177/15330338211016373.
Li L, Tang S, Yin JC, et al. Comprehensive Next-Generation Sequencing Reveals Novel Predictive Biomarkers of Recurrence and Thoracic Toxicity Risks After Chemoradiation Therapy in Limited Stage Small Cell Lung Cancer. Int J Radiat Oncol Biol Phys. 2022;112:1165–76. http://doi.org/10.1016/j.ijrobp.2021.12.009.
Niu L, Chu X, Yang X, et al. A multiomics approach-based prediction of radiation pneumonia in lung cancer patients: impact on survival outcome. J Cancer Res Clin Oncol. 2023;149:8923–34. http://doi.org/10.1007/s00432-023-04827-7.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Machine Learning-Based Model for Predicting Radiation Pneumonitis in Locally Advanced Non- Small Cell Lung Cancer Treated with IMRT-A Two-Centre Study

Status:

Version 1

Abstract

Figures

Introduction

Methods and Materials

Patients

Clinical and dosimetric data

CT image acquisition and radiomics features extraction

Evaluation of RP

Results

Patient characteristics and incidence of RP

Radiomics features selection

Development and validation of predictive models

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1