Electron Density and Biologically Effective Dose (BED) Radiomics-Based Machine Learning Models to Predict Late Radiation-Induced Subcutaneous Fibrosis

Breast cancer radiobiology: The renaissance of whole breast radiation fractionation (Review)

Breast cancer radiotherapy has evolved significantly, driven by decades of research into fractionation schedules aimed at optimizing treatment efficacy and minimizing toxicity. Initial trials such as NSABP B-06 and EBCTCG meta-analyses established the benefits of adjuvant whole-breast irradiation in reducing local recurrence and improving survival rates. The linear-quadratic (LQ) model provided a framework to understand tissue response to radiation, highlighting the importance of the α/β ratio in determining fractionation sensitivity. The present scoping review aimed to identify and describe hypofractionation regimens for whole breast radiotherapy and evaluate dose differences using the LQ model across proposed α/β ratios. A comprehensive PubMed search for clinical trials published since 2010 on hypo-fractionated regimens was performed. Studies discussing α/β ratios for breast cancer have been also searched. Data on dose, fractions and α/β ratios were collected, and biologically effective dose (BED) and equivalent dose in 2 Gy fractions were calculated. The coefficient of variation for BED varied with α/β ratios, showing the lowest variability for an α/β ratio of ~3 without tumor repopulation and increased with repopulation (BED-kT; k is a constant that depends on the repopulation rate of the tumor, and T is the total treatment time in days). Significant differences in BED variances were observed across α/β ratios (F-statistic 219.6, P<0.0001). START trials (P, A, and B) established α/β ratios of 3-4 Gy for breast cancer and normal tissues, confirming that hypofractionation is as effective as standard fractionation with potentially fewer late toxicities. Subsequent trials, such as FAST and FAST-Forward, demonstrated that ultra-hypofractionation is equivalent in tumor compared with conventional regimens. Further research is needed to gain a stronger understanding of radiobiological properties of breast cancer cells. Advances in radiotherapy technologies and the integration of biomarkers, radiomics and genomics are transforming treatment, moving towards precision medicine.

CT-based radiomics for predicting breast cancer radiotherapy side effects

Skin inflammation with the potential sequel of moist epitheliolysis and edema constitute the most frequent breast radiotherapy (RT) acute side effects. The aim of this study was to compare the predictive value of tissue-derived radiomics features to the total breast volume (TBV) for the moist cells epitheliolysis as a surrogate for skin inflammation, and edema. Radiomics features were extracted from computed tomography (CT) scans of 252 breast cancer patients from two volumes of interest: TBV and glandular tissue (GT). Machine learning classifiers were trained on radiomics and clinical features, which were evaluated for both side effects. The best radiomics model was a least absolute shrinkage and selection operator (LASSO) classifier, using TBV features, predicting moist cells epitheliolysis, achieving an area under the receiver operating characteristic (AUROC) of 0.74. This was comparable to TBV breast volume (AUROC of 0.75). Combined models of radiomics and clinical features did not improve performance. Exclusion of volume-correlated features slightly reduced the predictive performance (AUROC 0.71). We could demonstrate the general propensity of planning CT-based radiomics models to predict breast RT-dependent side effects. Mammary tissue was more predictive than glandular tissue. The radiomics features performance was influenced by their high correlation to TBV volume.

Comparing Performances of Predictive Models of Toxicity after Radiotherapy for Breast Cancer Using Different Machine Learning Approaches

PURPOSE

Different ML models were compared to predict toxicity in RT on a large cohort (n = 1314).

METHODS

The endpoint was RTOG G2/G3 acute toxicity, resulting in 204/1314 patients with the event. The dataset, including 25 clinical, anatomical, and dosimetric features, was split into 984 for training and 330 for internal tests. The dataset was standardized; features with a high p-value at univariate LR and with Spearman ρ>0.8 were excluded; synthesized data of the minority were generated to compensate for class imbalance. Twelve ML methods were considered. Model optimization and sequential backward selection were run to choose the best models with a parsimonious feature number. Finally, feature importance was derived for every model.

RESULTS

The model's performance was compared on a training-test dataset over different metrics: the best performance model was LightGBM. Logistic regression with three variables (LR3) selected via bootstrapping showed performances similar to the best-performing models. The AUC of test data is slightly above 0.65 for the best models (highest value: 0.662 with LightGBM).

CONCLUSIONS

No model performed the best for all metrics: more complex ML models had better performances; however, models with just three features showed performances comparable to the best models using many (n = 13-19) features.

Clinical Use of a Commercial Artificial Intelligence-Based Software for Autocontouring in Radiation Therapy: Geometric Performance and Dosimetric Impact

PURPOSE

When autocontouring based on artificial intelligence (AI) is used in the radiotherapy (RT) workflow, the contours are reviewed and eventually adjusted by a radiation oncologist before an RT treatment plan is generated, with the purpose of improving dosimetry and reducing both interobserver variability and time for contouring. The purpose of this study was to evaluate the results of application of a commercial AI-based autocontouring for RT, assessing both geometric accuracies and the influence on optimized dose from automatically generated contours after review by human operator.

MATERIALS AND METHODS

A commercial autocontouring system was applied to a retrospective database of 40 patients, of which 20 were treated with radiotherapy for prostate cancer (PCa) and 20 for head and neck cancer (HNC). Contours resulting from AI were compared against AI contours reviewed by human operator and human-only contours using Dice similarity coefficient (DSC), Hausdorff distance (HD), and relative volume difference (RVD). Dosimetric indices such as Dmean, D0.03cc, and normalized plan quality metrics were used to compare dose distributions from RT plans generated from structure sets contoured by humans assisted by AI against plans from manual contours. The reduction in contouring time obtained by using automated tools was also assessed. A Wilcoxon rank sum test was computed to assess the significance of differences. Interobserver variability of the comparison of manual vs. AI-assisted contours was also assessed among two radiation oncologists for PCa.

RESULTS

For PCa, AI-assisted segmentation showed good agreement with expert radiation oncologist structures with average DSC among patients ≥ 0.7 for all structures, and minimal radiation oncology adjustment of structures (DSC of adjusted versus AI structures ≥ 0.91). For HNC, results of comparison between manual and AI contouring varied considerably e.g., 0.77 for oral cavity and 0.11-0.13 for brachial plexus, but again, adjustment was generally minimal (DSC of adjusted against AI contours 0.97 for oral cavity, 0.92-0.93 for brachial plexus). The difference in dose for the target and organs at risk were not statistically significant between human and AI-assisted, with the only exceptions of D0.03cc to the anal canal and Dmean to the brachial plexus. The observed average differences in plan quality for PCa and HNC cases were 8% and 6.7%, respectively. The dose parameter changes due to interobserver variability in PCa were small, with the exception of the anal canal, where large dose variations were observed. The reduction in time required for contouring was 72% for PCa and 84% for HNC.

CONCLUSIONS

When an autocontouring system is used in combination with human review, the time of the RT workflow is significantly reduced without affecting dose distribution and plan quality.

Dosiomic-based prediction of dysgeusia in head & neck cancer patients treated with radiotherapy

PURPOSE

To investigate the potential of dosiomics in predicting radiotherapy-induced taste distortion (dysgeusia) in head & neck (H&N) cancer.

METHODS

A cohort of 80 H&N cancer patients treated with radical or adjuvant radiotherapy and with a follow-up of at least 24 months was enrolled. Treatment information, as well as tobacco and alcohol consumption were also collected. The whole tongue was manually delineated on the planning CT and mapped to the dose map retrieved from the treatment planning system. For every patient, 6 regions of the tongue were examined; for each of them, 145 dosiomic features were extracted from the dose map and fed to a logistic regression model to predict the grade of dysgeusia at follow-up, with and without including clinical features. A mean dose-based model was considered for reference.

RESULTS

Both dosiomics and mean dose models achieved good prediction performance for acute dysgeusia with AUC up to 0.88. For the dosiomic model, the central and anterior ⅔ regions of the tongue were the most predictive. For all models, a gradual reduction in the performance was observed at later times for chronic dysgeusia prediction, with higher values for dosiomics. The inclusion of smoke and alcohol habits did not improve model performances.

CONCLUSION

The dosiomic analysis of the dose to the tongue identified features able to predict acute dysgeusia. Dosiomics resulted superior to the conventional mean dose-based model for chronic dysgeusia prediction. Larger, prospective studies are needed to support these results before integrating dosiomics in radiotherapy planning.

Application of serum SERS technology based on thermally annealed silver nanoparticle composite substrate in breast cancer

Liquid biopsy is currently a non-destructive and convenient method of cancer screening, due to human blood containing a variety of cancer-related biomolecules. Therefore, the development of an accurate and rapid breast cancer screening technique combined with breast cancer serum is crucial for the treatment and prognosis of breast cancer patients. In this study, the surface enhanced Raman spectroscopy (SERS) technique is used to enhance the Raman spectroscopy (RS) signal of serum based on a high sensitivity thermally annealed silver nanoparticle/porous silicon bragg mirror (AgNPs/PSB) composite substrate. Compared with RS, SERS reflects more and stronger spectral peak information, which is beneficial to discover new biomarkers of breast cancer. At the same time, to further explore the diagnostic ability of SERS technology for breast cancer. In this study, the raw spectral data are processed by baseline correction, polynomial smoothing, and normalization. Then, the relevant feature information of SERS and RS is extracted by principal component analysis (PCA), and five classification models are established to compare the diagnostic performance of SERS and RS models respectively. The experimental results show that the breast cancer diagnosis model based on the improved SERS substrate combined with the machine learning algorithm can be used to distinguish breast cancer patients from controls. The accuracy, sensitivity, specificity and AUC values of the SVM model are 100%, 100%, 100% and 100%, respectively, as well as the training time of 4ms. The above experimental results show that the SERS technology based on AgNPs/PSB composite substrate, combined with machine learning methods, has great potential in the rapid and accurate identification of breast cancer patients.

A review of radiomics and genomics applications in cancers: the way towards precision medicine

The application of radiogenomics in oncology has great prospects in precision medicine. Radiogenomics combines large volumes of radiomic features from medical digital images, genetic data from high-throughput sequencing, and clinical-epidemiological data into mathematical modelling. The amalgamation of radiomics and genomics provides an approach to better study the molecular mechanism of tumour pathogenesis, as well as new evidence-supporting strategies to identify the characteristics of cancer patients, make clinical decisions by predicting prognosis, and improve the development of individualized treatment guidance. In this review, we summarized recent research on radiogenomics applications in solid cancers and presented the challenges impeding the adoption of radiomics in clinical practice. More standard guidelines are required to normalize radiomics into reproducible and convincible analyses and develop it as a mature field.

Predicting Local Failure after Partial Prostate Re-Irradiation Using a Dosiomic-Based Machine Learning Model

The aim of this study is to predict local failure after partial prostate re-irradiation for the treatment of isolated locally recurrent prostate cancer by using a machine learning classifier based on radiomic features from pre-treatment computed tomography (CT), positron-emission tomography (PET) and biological effective dose distribution (BED) of the radiotherapy plan. The analysis was conducted on a monocentric dataset of 43 patients with evidence of isolated intraprostatic recurrence of prostate cancer after primary external beam radiotherapy. All patients received partial prostate re-irradiation delivered by volumetric modulated arc therapy. The gross tumor volume (GTV) of each patient was manually contoured from planning CT, choline-PET and dose maps. An ensemble machine learning pipeline including unbalanced data correction and feature selection was trained using the radiomic and dosiomic features as input for predicting occurrence of local failure. The model performance was assessed using sensitivity, specificity, accuracy and area under receiver operating characteristic curves of the score function in 10-fold cross validation repeated 100 times. Local failure was observed in 13 patients (30%), with a median time to recurrence of 36.7 months (range = 6.1-102.4 months). A four variables ensemble machine learning model resulted in accuracy of 0.62 and AUC 0.65. According to our results, a dosiomic machine learning classifier can predict local failure after partial prostate re-irradiation.

Performance of Machine Learning and Texture Analysis for Predicting Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer with 3T MRI

Background: To evaluate the diagnostic performance of a Machine Learning (ML) algorithm based on Texture Analysis (TA) parameters in the prediction of Pathological Complete Response (pCR) to Neoadjuvant Chemoradiotherapy (nChRT) in Locally Advanced Rectal Cancer (LARC) patients. Methods: LARC patients were prospectively enrolled to undergo pre- and post-nChRT 3T MRI for initial loco-regional staging. TA was performed on axial T2-Weighted Images (T2-WI) to extract specific parameters, including skewness, kurtosis, entropy, and mean of positive pixels. For the assessment of TA parameter diagnostic performance, all patients underwent complete surgical resection, which served as a reference standard. ROC curve analysis was carried out to determine the discriminatory accuracy of each quantitative TA parameter to predict pCR. A ML-based decisional tree was implemented combining all TA parameters in order to improve diagnostic accuracy. Results: Forty patients were considered for final study population. Entropy, kurtosis and MPP showed statistically significant differences before and after nChRT in patients with pCR; in particular, when patients with Pathological Partial Response (pPR) and/or Pathological Non-Response (pNR) were considered, entropy and skewness showed significant differences before and after nChRT (all p < 0.05). In terms of absolute value changes, pre- and post-nChRT entropy, and kurtosis showed significant differences (0.31 ± 0.35, in pCR, −0.02 ± 1.28 in pPR/pNR, (p = 0.04); 1.87 ± 2.19, in pCR, −0.06 ± 3.78 in pPR/pNR (p = 0.0005); 107.91 ± 274.40, in pCR, −28.33 ± 202.91 in pPR/pNR, (p = 0.004), respectively). According to ROC curve analysis, pre-treatment kurtosis with an optimal cut-off value of ≤3.29 was defined as the best discriminative parameter, resulting in a sensitivity and specificity in predicting pCR of 81.5% and 61.5%, respectively. Conclusions: TA parameters extracted from T2-WI MRI images could play a key role as imaging biomarkers in the prediction of response to nChRT in LARC patients. ML algorithms can be used to efficiently combine all TA parameters in order to improve diagnostic accuracy.

A Novel Benchmarking Approach to Assess the Agreement among Radiomic Tools

Background The translation of radiomic models into clinical practice is hindered by the limited reproducibility of features across software and studies. Standardization is needed to accelerate this process and to bring radiomics closer to clinical deployment. Purpose To assess the standardization level of seven radiomic software programs and investigate software agreement as a function of built-in image preprocessing (eg, interpolation and discretization), feature aggregation methods, and the morphological characteristics (ie, volume and shape) of the region of interest (ROI). Materials and Methods The study was organized into two phases: In phase I, the two Image Biomarker Standardization Initiative (IBSI) phantoms were used to evaluate the IBSI compliance of seven software programs. In phase II, the reproducibility of all IBSI-standardized radiomic features across tools was assessed with two custom Italian multicenter Shared Understanding of Radiomic Extractors (ImSURE) digital phantoms that allowed, in conjunction with a systematic feature extraction, observations on whether and how feature matches between program pairs varied depending on the preprocessing steps, aggregation methods, and ROI characteristics. Results In phase I, the software programs showed different levels of completeness (ie, the number of computable IBSI benchmark values). However, the IBSI-compliance assessment revealed that they were all standardized in terms of feature implementation. When considering additional preprocessing steps, for each individual program, match percentages fell by up to 30%. In phase II, the ImSURE phantoms showed that software agreement was dependent on discretization and aggregation as well as on ROI shape and volume factors. Conclusion The agreement of radiomic software varied in relation to factors that had already been standardized (eg, interpolation and discretization methods) and factors that need standardization. Both dependences must be resolved to ensure the reproducibility of radiomic features and to pave the way toward the clinical adoption of radiomic models. © RSNA, 2022 Online supplemental material is available for this article. See also the editorial by Steiger in this issue.

Radiomic and Dosiomic Features for the Prediction of Radiation Pneumonitis Across Esophageal Cancer and Lung Cancer

PURPOSE

The aim was to investigate the advantages of dosiomic and radiomic features over traditional dose-volume histogram (DVH) features for predicting the development of radiation pneumonitis (RP), to validate the generalizability of dosiomic and radiomic features by using features selected from an esophageal cancer dataset and to use these features with a lung cancer dataset.

MATERIALS AND METHODS

A dataset containing 101 patients with esophageal cancer and 93 patients with lung cancer was included in this study. DVH and dosiomic features were extracted from 3D dose distributions. Radiomic features were extracted from pretreatment CT images. Feature selection was performed using only the esophageal cancer dataset. Four predictive models for RP (DVH, dosiomic, radiomic and dosiomic + radiomic models) were compared on the esophageal cancer dataset. We further used a lung cancer dataset for the external validation of the selected dosiomic and radiomic features from the esophageal cancer dataset. The performance of the predictive models was evaluated by the area under the curve (AUC) of the receiver operating characteristic curve (ROCAUC) and the AUC of the precision recall curve (PRAUC) metrics.

RESULT

The ROCAUCs and PRAUCs of the DVH, dosiomic, radiomic and dosiomic + radiomic models on esophageal cancer dataset were 0.67 ± 0.11 and 0.75 ± 0.10, 0.71 ± 0.10 and 0.77 ± 0.09, 0.71 ± 0.11 and 0.79 ± 0.09, and 0.75 ± 0.10 and 0.81 ± 0.09, respectively. The predictive performance of the dosiomic- and radiomic-based models was significantly higher than that of the DVH-based model with respect to esophageal cancer. The ROCAUCs and PRAUCs of the DVH, dosiomic, radiomic and dosiomic + radiomic models on the lung cancer dataset were 0.64 ± 0.18 and 0.37 ± 0.20, 0.67 ± 0.17 and 0.37 ± 0.20, 0.67 ± 0.16 and 0.45 ± 0.23, and 0.68 ± 0.16 and 0.44 ± 0.22, respectively. On the lung cancer dataset, the predictive performance of the radiomic and dosiomic + radiomic models was significantly higher than that of the DVH-based model. However, the PRAUC of the dosiomic-based model showed no significant difference relative to the corresponding RP prediction performance on the lung cancer dataset.

CONCLUSION

The results suggested that dosiomic and CT radiomic features could improve RP prediction in thoracic radiotherapy. Dosiomic and radiomic feature knowledge might be transferrable from esophageal cancer to lung cancer.

Combining computed tomography and biologically effective dose in radiomics and deep learning improves prediction of tumor response to robotic lung stereotactic body radiation therapy

PURPOSE

The aim of this study is to improve the performance of machine learning (ML) models in predicting response of non-small cell lung cancer (NSCLC) to stereotactic body radiation therapy (SBRT) by integrating image features from pre-treatment computed tomography (CT) with features from the biologically effective dose (BED) distribution.

MATERIALS AND METHODS

Image features, consisting of crafted radiomic features or machine-learned features extracted using a convolutional neural network, were calculated from pre-treatment CT data and from dose distributions converted into BED for 80 NSCLC lesions over 76 patients treated with robotic guided SBRT. ML models using different combinations of features were trained to predict complete or partial response according to response criteria in solid tumors, including radiomics CT (Rad_CT ), radiomics CT and BED (Rad_CT,BED ), deep learning (DL) CT (DL_CT ), and DL CT and BED (DL_CT,BED ). Training of ML included feature selection by neighborhood component analysis followed by ensemble ML using robust boosting. A model was considered as acceptable when the sum of average sensitivity and specificity on test data in repeated cross validations was at least 1.5.

RESULTS

Complete or partial response occurred in 58 out of 80 lesions. The best models to predict the tumor response were those using BED variables, achieving significantly better area under curve (AUC) and accuracy than those using only features from CT, including a Rad_CT,BED model using three radiomic features from BED, which scored an accuracy of 0.799 (95% confidence intervals (0.75-0.85)) and AUC of 0.773 (0.688-0.846), and a DL_CT,BED model also using three variables with an accuracy of 0.798 (0.649-0.829) and AUC of 0.812 (0.755-0.867).

CONCLUSION

According to our results, the inclusion of BED features improves the response prediction of ML models for lung cancer patients undergoing SBRT, regardless of the use of radiomic or DL features.

A Multicentre Evaluation of Dosiomics Features Reproducibility, Stability and Sensitivity

Simple Summary

Dosiomics is born directly as an extension of radiomics: it entails extracting features from the patients’ three-dimensional (3D) radiotherapy dose distribution rather than from conventional medical images to obtain specific spatial and statistical information. Dosiomic studies, in a multicentre setting, require assessing the features’ stability to dose calculation settings and the features’ capability in distinguishing different dose distributions. This study provides the first multicentre evaluation of the dosiomic features in terms of reproducibility, stability and sensitivity across various dose distributions obtained from multiple technologies and techniques and considering different dose calculation algorithms of TPS and two different resolutions of the dose grid. Harmonisation strategies to account for a possible variation in the dose distribution due to these confounding factors should be adopted when investigating a correlation between dosiomic features and clinical outcomes in multicentre studies.

Abstract

Dosiomics is a texture analysis method to produce dose features that encode the spatial 3D distribution of radiotherapy dose. Dosiomic studies, in a multicentre setting, require assessing the features’ stability to dose calculation settings and the features’ capability in distinguishing different dose distributions. Dose distributions were generated by eight Italian centres on a shared image dataset acquired on a dedicated phantom. Treatment planning protocols, in terms of planning target volume coverage and dose–volume constraints to the organs at risk, were shared among the centres to produce comparable dose distributions for measuring reproducibility/stability and sensitivity of dosiomic features. In addition, coefficient of variation (CV) was employed to evaluate the dosiomic features’ variation. We extracted 38,160 features from 30 different dose distributions from six regions of interest, grouped by four features’ families. A selected group of features (CV < 3 for the reproducibility/stability studies, CV > 1 for the sensitivity studies) were identified to support future multicentre studies, assuring both stable features when dose distributions variation is minimal and sensitive features when dose distribution variations need to be clearly identified. Dosiomic is a promising tool that could support multicentre studies, especially for predictive models, and encode the spatial and statistical characteristics of the 3D dose distribution.

Imaging-Based Staging of Hepatic Fibrosis in Patients with Hepatitis B: A Dynamic Radiomics Model Based on Gd-EOB-DTPA-Enhanced MRI

Accurate grading of liver fibrosis can effectively assess the severity of liver disease and help doctors make an appropriate diagnosis. This study aimed to perform the automatic staging of hepatic fibrosis on patients with hepatitis B, who underwent gadolinium ethoxybenzyl diethylenetriamine pentaacetic acid (Gd-EOB-DTPA)-enhanced magnetic resonance imaging with dynamic radiomics analysis. The proposed dynamic radiomics model combined imaging features from multi-phase dynamic contrast-enhanced (DCE) images and time-domain information. Imaging features were extracted from the deep learning-based segmented liver volume, and time-domain features were further explored to analyze the variation in features during contrast enhancement. Model construction and evaluation were based on a 132-case data set. The proposed model achieved remarkable performance in significant fibrosis (fibrosis stage S1 vs. S2–S4; accuracy (ACC) = 0.875, area under the curve (AUC) = 0.867), advanced fibrosis (S1–S2 vs. S3–S4; ACC = 0.825, AUC = 0.874), and cirrhosis (S1–S3 vs. S4; ACC = 0.850, AUC = 0.900) classifications in the test set. It was more dominant compared with the conventional single-phase or multi-phase DCE-based radiomics models, normalized liver enhancement, and some serological indicators. Time-domain features were found to play an important role in the classification models. The dynamic radiomics model can be applied for highly accurate automatic hepatic fibrosis staging.

Artificial Intelligence and the Medical Physicist: Welcome to the Machine

Artificial intelligence (AI) is a branch of computer science dedicated to giving machines or computers the ability to perform human-like cognitive functions, such as learning, problem-solving, and decision making. Since it is showing superior performance than well-trained human beings in many areas, such as image classification, object detection, speech recognition, and decision-making, AI is expected to change profoundly every area of science, including healthcare and the clinical application of physics to healthcare, referred to as medical physics. As a result, the Italian Association of Medical Physics (AIFM) has created the “AI for Medical Physics” (AI4MP) group with the aims of coordinating the efforts, facilitating the communication, and sharing of the knowledge on AI of the medical physicists (MPs) in Italy. The purpose of this review is to summarize the main applications of AI in medical physics, describe the skills of the MPs in research and clinical applications of AI, and define the major challenges of AI in healthcare.

A radiomics model of predicting tumor volume change of patients with stage III non-small cell lung cancer after radiotherapy

To predict the volume change of stage III NSCLC after radiotherapy with 60 Gy.This retrospective study included two independent cohorts, a train cohort of 192 patients, and a test cohort of 31 patients. We developed a radiomics model based on radiomics features and clinical variables. LIFEx package was used to extract radiomics texture features from CT images. The classification method was logistic regression analysis and feature selection was performed by correlation coefficients. Performance metrics of logistic regression include accuracy, precision, the receiver operating characteristic curves, and recall.The combination features of clinical variables and radiomics can predict the tumor volume change after radiotherapy with 88.7% accuracy (88.6% precision, 88.7% recall, and 88.7% ROC area).Radiomics features combined with medical knowledge have a great potential to predict accurately tumor volume change of stage III NSCLC after radiotherapy with 60 Gy.