External validation of radiomics-based predictive models in low-dose CT screening for early lung cancer diagnosis

CT-based radiomic prognostic vector (RPV) predicts survival and stromal histology in high-grade serous ovarian cancer: an external validation study

OBJECTIVES

In women with high-grade serous ovarian cancer (HGSOC), a CT-based radiomic prognostic vector (RPV) predicted stromal phenotype and survival after primary surgery. The study's purpose was to fully externally validate RPV and its biological correlate.

MATERIALS AND METHODS

In this retrospective study, ovarian masses on CT scans of HGSOC patients, who underwent primary cytoreductive surgery in an ESGO-certified Center between 2002 and 2017, were segmented for external RPV score calculation and then correlated with overall survival (OS) and progression-free survival (PFS). A subset of tissue samples subjected to fibronectin immunohistochemistry were evaluated by a gynaeco-pathologist for stromal content. Kaplan-Meier log-rank test and a Cox proportional hazards model were used for outcome analysis.

RESULTS

Among 340 women with HGSOC, 244 ovarian lesions were available for segmentation in 198 women (mean age 59.8 years, range 34-92). Median OS was 48.69 months (IQR: 27.0-102.5) and PFS was 19.3 months (IQR: 13-32.2). Using multivariate Cox analysis, poor OS was associated with RPV-high (HR 3.17; 95% CI: 1.32-7.60; p = 0.0099), post-operative residual disease (HR 2.04; 95% CI: 1.30-3.20; p = 0.0020), and FIGO stage III/IV (HR 1.79; 95% CI: 1.11-2.86; p = 0.016). Age did not influence OS. RPV-high tissue had higher stromal content based on fibronectin expression (mean 48.9%, SD 10.5%) compared to RPV-low cases (mean 14.9%, SD 10.5%, p < 0.0001). RPV score was not significantly associated with PFS.

CONCLUSION

Patients with HGSOC and RPV-high ovarian mass on pre-operative CT had significantly worse OS following primary surgery and a higher stromal content compared to RPV-low masses, externally validating the RPV and its biological interpretation.

KEY POINTS

Question Can the performance of a previously described RPV in women with HGSOC be replicated when licenced to an external institution? Findings External validation of RPV among 244 ovarian lesions demonstrated that, on multivariate analysis, OS was associated with RPV, stage, and postoperative residual disease, replicating previous findings. Clinical relevance External validation of a radiomic tool is an essential step in translation to clinical applicability and provides the basis for prospective validation. In clinical practice, this RPV may allow more personalized decision-making for women with ovarian cancer being considered for extensive cytoreductive surgery.

The influence of image selection and segmentation on the extraction of lung cancer imaging radiomics features using 3D-Slicer software

PURPOSE

Extracting image features can predict the prognosis and treatment effect of non-small cell lung cancer, which has been increasingly confirmed. However, the specific operation using 3D-Slicer still lacks standardization. For example, image segmentation is manually performed based on the lung window or automatically performed through the mediastinal window. The images used for feature extraction are either enhanced or plain scanned. It is questionable whether these influencing factors will affect the extraction results and which results will be affected. This article intends to preliminarily explore the above issues.

METHODS

This article downloaded images of 22 patients with lung cancer from The Cancer Imaging Archive (TCIA), including 11 cases of adenocarcinoma and 11 cases of squamous cell carcinoma. Perform tumor image segmentation on the lung window and mediastinal window of the plain scan image, and the lung window and mediastinal window of the enhanced image. Manual drawing is used on the lung window, and automatic drawing is used on the mediastinal window and make manual modifications. Extracting radiomics features using Python radiomics. Firstly, analyze the image features of the original sequence and perform the Shapiro test. If it follows a normal distribution, perform an analysis of variance. If it does not follow a normal distribution, perform the Friedman test. Compare the significantly different image features pairwise. Then, a preliminary analysis was conducted on the differences between squamous cell carcinoma and adenocarcinoma in each group.

RESULTS

A total of 88 sets of imaging features were extracted, with 107 features in each group. Among them, 33 features showed significant differences. Continuing with pairwise repeated testing, it was found that there were 2 significant differences between enhanced and plain lung windows. There were 12 significant differences between enhanced lung windows and plain mediastinal windows. There is one significant difference between plain scanning and enhancement mediastinal window. There are 14 significant differences between the plain lung window and the enhanced mediastinal window groups. There are 14 significant differences between the lung window and the mediastinal window in the plain scan. There are 13 significant differences between the enhanced lung window and the mediastinal window. According to pathological grouping testing, it was found that there 54 significant differences between squamous cell carcinoma and adenocarcinoma.

CONCLUSION

The enhancement of lung CT has a relatively small impact on extracting image features, while selecting lung or mediastinal windows for image segmentation has a significant impact on extracting image features. Therefore, choosing lung or mediastinal windows for feature extraction should be carefully considered, as the size of the image segmentation range has a significant impact on image features. The impact of lung squamous cell carcinoma and adenocarcinoma on imaging features is also significant, indicating a high possibility of distinguishing between squamous cell carcinoma and adenocarcinoma based on radiomics (Liu C, He Y, Luo J, The Influence of Image Selection and Segmentation on the Extraction of Lung Cancer Imaging Radiomics Features Using 3D-Slicer Software, 2024).

Evaluation of radiomic feature harmonization techniques for benign and malignant pulmonary nodules

Background

Conventional methods for detecting lung cancer early are often qualitative and subject to interpretation. Radiomics provides quantitative characteristics of pulmonary nodules (PNs) in medical images, but variability in medical image acquisition is an obstacle to consistent clinical application of these quantitative features. Correcting radiomic features' dependency on acquisition parameters is problematic when combining data from benign and malignant PNs, as is necessary when the goal is to diagnose lung cancer, because acquisition effects may differ between them due to their biological differences.

Purpose

We evaluated whether we must account for biological differences between benign and malignant PNs when correcting the dependency of radiomic features on acquisition parameters, and we compared methods of doing this using ComBat harmonization.

Methods

This study used a dataset of 567 clinical chest CT scans containing both malignant and benign PNs. Scans were grouped as benign, malignant, or lung cancer screening (mixed benign and malignant). Preprocessing and feature extraction from ROIs were performed using PyRadiomics. Optimized Permutation Nested ComBat harmonization was performed on extracted features to account for variability in four imaging protocols: contrast enhancement, scanner manufacturer, acquisition voltage, focal spot size. Three methods were compared: harmonizing all data collectively in the standard manner, harmonizing all data with a covariate to preserve distinctions between subgroups, and harmonizing subgroups separately. A significant (p ≤ 0.05) Kruskal-Wallis test determined whether harmonization removed a feature's dependency on an acquisition parameter. A LASSO-SVM pipeline was trained using acquisition-independent radiomic features to predict whether PNs were malignant or benign. To evaluate the predictive information made available by each harmonization method, the trained harmonization estimators and predictive model were applied to a corresponding unseen test set. Harmonization and predictive performance metrics were assessed over 10 trials of 5-fold cross validation.

Results

Kruskal-Wallis defined an average 2.1% of features (95% CI: 1.9-2.4%) as acquisition-independent when data were harmonized collectively, 27.3% of features (95% CI: 25.7-28.9%) as acquisition-independent when harmonized with a covariate, and 90.9% of features (95% CI: 90.4-91.5%) as acquisition-independent when harmonized separately. LASSO-SVM models trained on data harmonized separately or with a covariate had higher ROC-AUC for lung cancer screening scans than models trained on data harmonized without distinction between benign and malignant tissues (Delong test, Holm-Bonferroni adjusted p ≤ 0.05). There was not a conclusive difference in ROC-AUC between models trained on data harmonized separately and models trained on data harmonized with a covariate.

Conclusions

Radiomic features of benign and malignant PNs require different corrective transformations to recover acquisition-independent distributions. This can be done using separate harmonization or harmonization with a covariate. Separate harmonization enabled the greatest number of predictive features to be used in a machine learning model to retrospectively detect lung cancer. Features harmonized separately and features harmonized with a covariate enabled predictive models to achieve similar performance on lung cancer screening scans.

Lung nodule classification using radiomics model trained on degraded SDCT images

BACKGROUND AND OBJECTIVE

Low-dose computed tomography (LDCT) screening has shown promise in reducing lung cancer mortality; however, it suffers from high false positive rates and a scarcity of available annotated datasets. To overcome these challenges, we propose a novel approach using synthetic LDCT images generated from standard-dose CT (SDCT) scans from the LIDC-IDRI dataset. Our objective is to develop and validate an interpretable radiomics-based model for distinguishing likely benign from likely malignant pulmonary nodules.

METHODS

From a total of 1010 CT images (695 SDCTs and 315 LDCTs), we degraded SDCTs in the sinogram domain and obtained 1950 nodules as the training set. The 675 nodules from the LDCTs were stratified into 50%-50% partitions for validation and testing. Radiomic features were extracted from nodules, and three feature sets were assessed using: a) only shape and size (SS) features, b) all features but SS features, and c) all features. A systematic pipeline was developed to optimize the feature set and evaluate multiple machine learning models. Models were trained using degraded SDCT, validated and tested on the LDCT nodules.

RESULTS

Training a logistic regression model using three SS features yielded the most promising results, achieving on the test set mean balanced accuracy, sensitivity, specificity, and AUC-ROC scores of 0.81, 0.76, 0.85, and 0.87, respectively.

CONCLUSIONS

Our study demonstrates the feasibility and effectiveness of using synthetic LDCT images for developing a relatively accurate radiomics-based model in lung nodule classification. This approach addresses challenges associated with LDCT screening, offering potential implications for improving lung cancer detection and reducing false positives.

Decoding Radiomics: A Step-by-Step Guide to Machine Learning Workflow in Hand-Crafted and Deep Learning Radiomics Studies

Although radiomics research has experienced rapid growth in recent years, with numerous studies dedicated to the automated extraction of diagnostic and prognostic information from various imaging modalities, such as CT, PET, and MRI, only a small fraction of these findings has successfully transitioned into clinical practice. This gap is primarily due to the significant methodological challenges involved in radiomics research, which emphasize the need for a rigorous evaluation of study quality. While many technical aspects may lie outside the expertise of most radiologists, having a foundational knowledge is essential for evaluating the quality of radiomics workflows and contributing, together with data scientists, to the development of models with a real-world clinical impact. This review is designed for the new generation of radiologists, who may not have specialized training in machine learning or radiomics, but will inevitably play a role in this evolving field. The paper has two primary objectives: first, to provide a clear, systematic guide to radiomics study pipeline, including study design, image preprocessing, feature selection, model training and validation, and performance evaluation. Furthermore, given the critical importance of evaluating the robustness of radiomics studies, this review offers a step-by-step guide to the application of the METhodological RadiomICs Score (METRICS, 2024)-a newly proposed tool for assessing the quality of radiomics studies. This roadmap aims to support researchers and reviewers alike, regardless of their machine learning expertise, in utilizing this tool for effective study evaluation.

radMLBench: A dataset collection for benchmarking in radiomics

BACKGROUND

New machine learning methods and techniques are frequently introduced in radiomics, but they are often tested on a single dataset, which makes it challenging to assess their true benefit. Currently, there is a lack of a larger, publicly accessible dataset collection on which such assessments could be performed. In this study, a collection of radiomics datasets with binary outcomes in tabular form was curated to allow benchmarking of machine learning methods and techniques.

METHODS

A variety of journals and online sources were searched to identify tabular radiomics data with binary outcomes, which were then compiled into a homogeneous data collection that is easily accessible via Python. To illustrate the utility of the dataset collection, it was applied to investigate whether feature decorrelation prior to feature selection could improve predictive performance in a radiomics pipeline.

RESULTS

A total of 50 radiomic datasets were collected, with sample sizes ranging from 51 to 969 and 101 to 11165 features. Using this data, it was observed that decorrelating features did not yield any significant improvement on average.

CONCLUSIONS

A large collection of datasets, easily accessible via Python, suitable for benchmarking and evaluating new machine learning techniques and methods was curated. Its utility was exemplified by demonstrating that feature decorrelation prior to feature selection does not, on average, lead to significant performance gains and could be omitted, thereby increasing the robustness and reliability of the radiomics pipeline.

Effect of Deep Learning Image Reconstruction Algorithms on Radiomic Features of Pulmonary Nodules in Ultra-Low-Dose CT

OBJECTIVE

The purpose of this study is to explore the impact of deep learning image reconstruction (DLIR) algorithm on the quantification of radiomic features in ultra-low-dose computed tomography (ULD-CT) compared with adaptive statistical iterative reconstruction-Veo (ASIR-V).

METHODS

One hundred eighty-three patients with pulmonary nodules underwent standard-dose computed tomography (SDCT) (4.30 ± 0.36 mSv) and ULD-CT (UL-A, 0.57 ± 0.09 mSv or UL-B, 0.33 ± 0.04 mSv). SDCT was the reference standard using (ASIR-V) at 50% strength (50%ASIR-V). ULD-CT was reconstructed with 50%ASIR-V, DLIR at medium and high strength (DLIR-M, DLIR-H). Radiomics analysis extracted 102 features, and the intraclass correlation coefficient (ICC) quantified reproducibility between ULD-CT and SDCT reconstructed by 50%ASIR-V, DLIR-M, and DLIR-H for each feature.

RESULTS

Among 102 radiomic features, the percentages of reproducibility of 50%ASIR-V, DLIR-M, and DLIR-H were 48.04% (49/102), 49.02% (50/102), and 52.94% (54/102), respectively. Shape and first order features demonstrated high reproducibility across different reconstruction algorithms and radiation doses, with mean ICC values exceeding 0.75. In texture features, DLIR-M and DLIR-H showed improved mean ICC values for pure ground glass nodules (pGGNs) from 0.69 ± 0.23 to 0.75 ± 0.18 and 0.81 ± 0.12, respectively, compared with 50%ASIR-V. Similarly, the mean ICC values for solid nodules (SNs) increased from 0.60 ± 0.19 to 0.66 ± 0.14 and 0.69 ± 0.13, respectively. Additionally, the mean ICC values of texture features for pGGNs and SNs in both ULD-CT groups decreased with reduced radiation dose.

CONCLUSIONS

DLIR can improve the reproducibility of radiomic features at ultra-low doses compared with ASIR-V. In addition, pGGNs showed better reproducibility at ultra-low doses than SNs.

Addressing intra- and inter-institution variability of a radiomic framework based on Apparent Diffusion Coefficient in prostate cancer

BACKGROUND

Prostate cancer (PCa) is a highly heterogeneous disease, making tailored treatment approaches challenging. Magnetic resonance imaging (MRI), notably diffusion-weighted imaging (DWI) and the derived Apparent Diffusion Coefficient (ADC) maps, plays a crucial role in PCa characterization. In this context, radiomics is a very promising approach able to disclose insights from MRI data. However, the sensitivity of radiomic features to MRI settings, encompassing DWI protocols and multicenter variations, requires the development of robust and generalizable models.

PURPOSE

To develop a comprehensive radiomics framework for noninvasive PCa characterization using ADC maps, focusing on identifying reliable imaging biomarkers against intra- and inter-institution variations.

MATERIALS AND METHODS

Two patient cohorts, including an internal cohort (118 PCa patients) used for both training (75%) and hold-out testing (25%), and an external cohort (50 PCa patients) for independent testing, were employed in the study. DWI images were acquired with three different DWI protocols on two different MRI scanners: two DWI protocols acquired on a 1.5-T scanner for the internal cohort, and one DWI protocol acquired on a 3-T scanner for the external cohort. One hundred and seven radiomics features (i.e., shape, first order, texture) were extracted from ADC maps of the whole prostate gland. To address variations in DWI protocols and multicenter variability, a dedicated pipeline, including two-way ANOVA, sequential-feature-selection (SFS), and ComBat features harmonization was implemented. Mann-Whitney U-tests (α = 0.05) were performed to find statistically significant features dividing patients with different tumor characteristics in terms of Gleason score (GS) and T-stage. Support-Vector-Machine models were then developed to predict GS and T-stage, and the performance was assessed through the area under the curve (AUC) of receiver-operating-characteristic curves.

RESULTS

Downstream of ANOVA, two subsets of 38 and 41 features stable against DWI protocol were identified for GS and T-stage, respectively. Among these, SFS revealed the most predictive features, yielding an AUC of 0.75 (GS) and 0.70 (T-stage) in the hold-out test. Employing ComBat harmonization improved the external-test performance of the GS model, raising AUC from 0.72 to 0.78.

CONCLUSION

By incorporating stable features with a harmonization procedure and validating the model on an external dataset, model robustness, and generalizability were assessed, highlighting the potential of ADC and radiomics for PCa characterization.

Differential Radiomics-Based Signature Predicts Lung Cancer Risk Accounting for Imaging Parameters in NLST Cohort

OBJECTIVE

Lung cancer remains the leading cause of cancer-related mortality worldwide, with most cases diagnosed at advanced stages. Hence, there is a need to develop effective predictive models for early detection. This study aims to investigate the impact of imaging parameters and delta radiomic features from temporal scans on lung cancer risk prediction.

METHODS

Using the National Lung Screening Trial (NLST) within a nested case-control study involving 462 positive screenings, radiomic features were extracted from temporal computed tomography (CT) scans and harmonized with ComBat method to adjust variations in slice thickness category (TC) and reconstruction kernel type (KT). Both harmonized and non-harmonized features from baseline (T0), delta features between T0 and a year later (T1), and combined T0 and delta features were utilized for the analysis. Feature reduction was done using LASSO, followed by five feature selection (FS) methods and nine machine learning (ML) models, evaluated with 5-fold cross-validation repeated 10 times. Synthetic Minority Oversampling Technique (SMOTE) was applied to address class imbalances for lung cancer risk prediction.

RESULTS

Models using delta features outperformed baseline features, with SMOTE consistently boosting performance when using combination of baseline and delta features. TC-based harmonized features improved performance with SMOTE, but overall, harmonization did not significantly enhance the model performance. The highest test score of 0.76 was achieved in three scenarios: delta features with a Gradient Boosting (GB) model (TC-based harmonization and MultiSurf FS); and T0 + delta features, with both a Support Vector Classifier (SVC) model (KT-based harmonization and F-test FS), and an XGBoost (XGB) model (TC-based harmonization and Mutual Information (MI) FS), all using SMOTE.

CONCLUSIONS

This study underscores the significance of delta radiomic features and balanced datasets to improve lung cancer prediction. While our findings are based on a subsample of NLST data, they provide a valuable foundation for further exploration. Further research is needed to assess the impact of harmonization on imaging-derived models. Future investigations should explore advanced harmonization techniques and additional imaging parameters to develop robust radiomics-based biomarkers of lung cancer risk.

Predicting Postoperative Length of Stay in Patients Undergoing Laparoscopic Right Hemicolectomy for Colon Cancer: A Machine Learning Approach Using SICE (Società Italiana di Chirurgia Endoscopica) CoDIG Data

The evolution of laparoscopic right hemicolectomy, particularly with complete mesocolic excision (CME) and central vascular ligation (CVL), represents a significant advancement in colon cancer surgery. The CoDIG 1 and CoDIG 2 studies highlighted Italy's progressive approach, providing useful findings for optimizing patient outcomes and procedural efficiency. Within this context, accurately predicting postoperative length of stay (LoS) is crucial for improving resource allocation and patient care, yet its determination through machine learning techniques (MLTs) remains underexplored. This study aimed to harness MLTs to forecast the LoS for patients undergoing right hemicolectomy for colon cancer, using data from the CoDIG 1 (1224 patients) and CoDIG 2 (788 patients) studies. Multiple MLT algorithms, including random forest (RF) and support vector machine (SVM), were trained to predict LoS, with CoDIG 1 data used for internal validation and CoDIG 2 data for external validation. The RF algorithm showed a strong internal validation performance, achieving the best performances and a 0.92 ROC in predicting long-term stays (more than 5 days). External validation using the SVM model demonstrated 75% ROC values. Factors such as fast-track protocols, anastomosis, and drainage emerged as key predictors of LoS. Integrating MLTs into predicting postoperative LOS in colon cancer surgery offers a promising avenue for personalized patient care and improved surgical management. Using intraoperative features in the algorithm enables the profiling of a patient's stay based on the planned intervention. This issue is important for tailoring postoperative care to individual patients and for hospitals to effectively plan and manage long-term stays for more critical procedures.

Construction of a predictive model of 2-3 cm ground-glass nodules developing into invasive lung adenocarcinoma using high-resolution CT

Background

The purpose of this study was to analyze the imaging risk factors for the development of 2-3 cm ground-glass nodules (GGN) for invasive lung adenocarcinoma and to establish a nomogram prediction model to provide a reference for the pathological prediction of 2-3 cm GGN and the selection of surgical procedures.

Methods

We reviewed the demographic, imaging, and pathological information of 596 adult patients who underwent 2-3 cm GGN resection, between 2018 and 2022, in the Department of Thoracic Surgery, Second Affiliated Hospital of the Air Force Medical University. Based on single factor analysis, the regression method was used to analyze multiple factors, and a nomogram prediction model for 2-3 cm GGN was established.

Results

(1) The risk factors for the development of 2-3 cm GGN during the invasion stage of the lung adenocarcinoma were pleural depression sign (OR = 1.687, 95%CI: 1.010-2.820), vacuole (OR = 2.334, 95%CI: 1.222-4.460), burr sign (OR = 2.617, 95%CI: 1.008-6.795), lobulated sign (OR = 3.006, 95%CI: 1.098-8.227), bronchial sign (OR = 3.134, 95%CI: 1.556-6.310), diameter of GGN (OR = 3.118, 95%CI: 1.151-8.445), and CTR (OR = 172.517, 95%CI: 48.023-619.745). (2) The 2-3 cm GGN risk prediction model was developed based on the risk factors with an AUC of 0.839; the calibration curve Y was close to the X-line, and the decision curve was drawn in the range of 0.0-1.0.

Conclusion

We analyzed the risk factors for the development of 2-3  cm GGN during the invasion stage of the lung adenocarcinoma. The predictive model developed based on the above factors had some clinical significance.

Novel Functional Radiomics for Prediction of Cardiac Positron Emission Tomography Avidity in Lung Cancer Radiotherapy

PURPOSE

Traditional methods of evaluating cardiotoxicity focus on radiation doses to the heart. Functional imaging has the potential to provide improved prediction for cardiotoxicity for patients with lung cancer. Fluorine-18 (18F) fluorodeoxyglucose (FDG)-positron emission tomography (PET)/computed tomography (CT) imaging is routinely obtained in a standard cancer staging workup. This work aimed to develop a radiomics model predicting clinical cardiac assessment using 18F-FDG PET/CT scans before thoracic radiation therapy.

METHODS

Pretreatment 18F-FDG PET/CT scans from three study populations (N = 100, N = 39, N = 70) were used, comprising two single-institutional protocols and one publicly available data set. A clinician (V.J.) classified the PET/CT scans per clinical cardiac guidelines as no uptake, diffuse uptake, or focal uptake. The heart was delineated, and 210 novel functional radiomics features were selected to classify cardiac FDG uptake patterns. Training data were divided into training (80%)/validation (20%) sets. Feature reduction was performed using the Wilcoxon test, hierarchical clustering, and recursive feature elimination. Ten-fold cross-validation was carried out for training, and the accuracy of the models to predict clinical cardiac assessment was reported.

RESULTS

From 202 of 209 scans, cardiac FDG uptake was scored as no uptake (39.6%), diffuse uptake (25.3%), and focal uptake (35.1%), respectively. Sixty-two independent radiomics features were reduced to nine clinically pertinent features. The best model showed 93% predictive accuracy in the training data set and 80% and 92% predictive accuracy in two external validation data sets.

CONCLUSION

This work used an extensive patient data set to develop a functional cardiac radiomic model from standard-of-care 18F-FDG PET/CT scans, showing good predictive accuracy. The radiomics model has the potential to provide an automated method to predict existing cardiac conditions and provide an early functional biomarker to identify patients at risk of developing cardiac complications after radiotherapy.

A machine learning-based sonomics for prediction of thyroid nodule malignancies

OBJECTIVES

This study aims to use ultrasound derived features as biomarkers to assess the malignancy of thyroid nodules in patients who were candidates for FNA according to the ACR TI-RADS guidelines.

METHODS

Two hundred and ten patients who met the selection criteria were enrolled in the study and subjected to ultrasound-guided FNA of thyroid nodules. Different radiomics features were extracted from sonographic images, including intensity, shape, and texture feature sets. Least Absolute Shrinkage and Selection Operator (LASSO), Minimum Redundancy Maximum Relevance (MRMR), and Random Forests/Extreme Gradient Boosting Machine (XGBoost) algorithms were used for feature selection and classification of the univariate and multivariate modeling, respectively. Evaluation of models performed using accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC).

RESULTS

In the univariate analysis, Gray Level Run Length Matrix - Run-Length Non-Uniformity (GLRLM-RLNU) and gray-level zone length matrix - Run-Length Non-Uniformity (GLZLM-GLNU) (both with an AUC of 0.67) were top-performing for predicting nodules malignancy. In the multivariate analysis of the training dataset, the AUC of all combinations of feature selection algorithms and classifiers was 0.99, and the highest sensitivity was for XGBoost classifier and MRMR feature selection algorithms (0.99). Finally, the test dataset was used to evaluate our model in which XGBoost classifier with MRMR and LASSO feature selection algorithms had the highest performance (AUC = 0.95).

CONCLUSIONS

Ultrasound-extracted features can be used as non-invasive biomarkers for thyroid nodules' malignancy prediction.

Dual-Centre Harmonised Multimodal Positron Emission Tomography/Computed Tomography Image Radiomic Features and Machine Learning Algorithms for Non-small Cell Lung Cancer Histopathological Subtype Phenotype Decoding

AIMS

We aimed to build radiomic models for classifying non-small cell lung cancer (NSCLC) histopathological subtypes through a dual-centre dataset and comprehensively evaluate the effect of ComBat harmonisation on the performance of single- and multimodality radiomic models.

MATERIALS AND METHODS

A public dataset of NSCLC patients from two independent centres was used. Two image fusion methods, namely guided filtering-based fusion and image fusion based on visual saliency map and weighted least square optimisation, were used. Radiomic features were extracted from each scan, including first-order, texture and moment-invariant features. Subsequently, ComBat harmonisation was applied to the extracted features from computed tomography (CT), positron emission tomography (PET) and fused images to correct the centre effect. For feature selection, least absolute shrinkage and selection operator (Lasso) and recursive feature elimination (RFE) were investigated. For machine learning, logistic regression (LR), support vector machine (SVM) and AdaBoost were evaluated for classifying NSCLC subtypes. Training and evaluation of the models were carried out in a robust framework to offset plausible errors and performance was reported using area under the curve, balanced accuracy, sensitivity and specificity before and after harmonisation. N-way ANOVA was used to assess the effect of different factors on the performance of the models.

RESULTS

Support vector machine fed with selected features by recursive feature elimination from a harmonised PET feature set achieved the highest performance (area under the curve = 0.82) in classifying NSCLC histopathological subtypes. Although the performance of the models did not significantly improve for CT images after harmonisation, the performance of PET and guided filtering-based fusion feature signatures significantly improved for almost all models. Although the selection of the image modality and feature selection methods was effective on the performance of the model (ANOVA P-values <0.001), machine learning and harmonisation did not change the performance significantly (ANOVA P-values = 0.839 and 0.292, respectively).

CONCLUSION

This study confirmed the potential of radiomic analysis on PET, CT and hybrid images for histopathological classification of NSCLC subtypes.

Artificial intelligence applications in computed tomography in gastric cancer: a narrative review

Background and Objective

Artificial intelligence (AI) is a revolutionary technique which is deeply impacting and reshaping clinical practice in oncology. This review aims to summarize the current status of the clinical application of AI-based computed tomography (CT) for gastric cancer (GC), focusing on diagnosis, genetic status detection and risk prediction of metastasis, prognosis and treatment efficacy. The challenges and prospects for future research will also be discussed.

Methods

We searched the PubMed/MEDLINE database to identify clinical studies published between 1990 and November 2022 that investigated AI applications in CT in GC. The major findings of the verified studies were summarized.

Key Content and Findings

AI applications in CT images have attracted considerable attention in various fields such as diagnosis, prediction of metastasis risk, survival, and treatment response. These emerging techniques have shown a high potential to outperform clinicians in diagnostic accuracy and time-saving.

Conclusions

AI-powered tools showed great potential to increase diagnostic accuracy and reduce radiologists' workload. However, the goal of AI is not to replace human ability but to help oncologists make decisions in their practice. Therefore, radiologists should play a predominant role in AI applications and decide the best ways to integrate these complementary techniques within clinical practice.

CT-based radiomics prediction of complete response after stereotactic body radiation therapy for patients with lung metastases

PURPOSE

Stereotactic body radiotherapy (SBRT) is a key treatment modality for lung cancer patients. This study aims to develop a machine learning-based prediction model of complete response for lung oligometastatic cancer patients undergoing SBRT.

MATERIALS AND METHODS

CT images of 80 pulmonary oligometastases from 56 patients treated with SBRT were analyzed. The gross tumor volumes (GTV) were contoured on CT images. Patients that achieved complete response (CR) at 4 months were defined as responders. For each GTV, 107 radiomic features were extracted using the Pyradiomics software. The concordance correlation coefficients (CCC) between the region of interest (ROI)-based radiomics features obtained by the two segmentations were calculated. Pairwise feature interdependencies were evaluated using the Spearman rank correlation coefficient. The association of clinical variables and radiomics features with CR was evaluated with univariate logistic regression. Two supervised machine learning models, the logistic regression (LR) and the classification and regression tree analysis (CART), were trained to predict CR. The models were cross-validated using a five-fold cross-validation. The performance of models was assessed by receiver operating characteristic curve (ROC) and class-specific accuracy, precision, recall, and F1-measure evaluation metrics.

RESULTS

Complete response was associated with four radiomics features, namely the surface to volume ratio (SVR; p = 0.003), the skewness (Skew; p = 0.027), the correlation (Corr; p = 0.024), and the grey normalized level uniformity (GNLU; p = 0.015). No significant relationship between clinical parameters and CR was found. In the validation set, the developed LR and CART machine learning models had an accuracy, precision, and recall of 0.644 and 0.750, 0.644 and 0.651, and 0.635 and 0.754, respectively. The area under the curve for CR prediction was 0.707 and 0.753 for the LR and CART models, respectively.

CONCLUSION

This analysis demonstrates that radiomics features obtained from pretreatment CT could predict complete response of lung oligometastases following SBRT.

Impact of localized fine tuning in the performance of segmentation and classification of lung nodules from computed tomography scans using deep learning

Background

Algorithm malfunction may occur when there is a performance mismatch between the dataset with which it was developed and the dataset on which it was deployed.

Methods

A baseline segmentation algorithm and a baseline classification algorithm were developed using public dataset of Lung Image Database Consortium to detect benign and malignant nodules, and two additional external datasets (i.e., HB and XZ) including 542 cases and 486 cases were involved for the independent validation of these two algorithms. To explore the impact of localized fine tuning on the individual segmentation and classification process, the baseline algorithms were fine tuned with CT scans of HB and XZ datasets, respectively, and the performance of the fine tuned algorithms was tested to compare with the baseline algorithms.

Results

The proposed baseline algorithms of both segmentation and classification experienced a drop when directly deployed in external HB and XZ datasets. Comparing with the baseline validation results in nodule segmentation, the fine tuned segmentation algorithm obtained better performance in Dice coefficient, Intersection over Union, and Average Surface Distance in HB dataset (0.593 vs. 0.444; 0.450 vs. 0.348; 0.283 vs. 0.304) and XZ dataset (0.601 vs. 0.486; 0.482 vs. 0.378; 0.225 vs. 0.358). Similarly, comparing with the baseline validation results in benign and malignant nodule classification, the fine tuned classification algorithm had improved area under the receiver operating characteristic curve value, accuracy, and F1 score in HB dataset (0.851 vs. 0.812; 0.813 vs. 0.769; 0.852 vs. 0.822) and XZ dataset (0.724 vs. 0.668; 0.696 vs. 0.617; 0.737 vs. 0.668).

Conclusions

The external validation performance of localized fine tuned algorithms outperformed the baseline algorithms in both segmentation process and classification process, which showed that localized fine tuning may be an effective way to enable a baseline algorithm generalize to site-specific use.

Clinical Scores, Biomarkers and IT Tools in Lung Cancer Screening-Can an Integrated Approach Overcome Current Challenges?

As most lung cancer (LC) cases are still detected at advanced and incurable stages, there are increasing efforts to foster detection at earlier stages by low dose computed tomography (LDCT) based LC screening. In this scoping review, we describe current advances in candidate selection for screening (selection phase), technical aspects (screening), and probability evaluation of malignancy of CT-detected pulmonary nodules (PN management). Literature was non-systematically assessed and reviewed for suitability by the authors. For the selection phase, we describe current eligibility criteria for screening, along with their limitations and potential refinements through advanced clinical scores and biomarker assessments. For LC screening, we discuss how the accuracy of computerized tomography (CT) scan reading might be augmented by IT tools, helping radiologists to cope with increasing workloads. For PN management, we evaluate the precision of follow-up scans by semi-automatic volume measurements of CT-detected PN. Moreover, we present an integrative approach to evaluate the probability of PN malignancy to enable safe decisions on further management. As a clear limitation, additional validation studies are required for most innovative diagnostic approaches presented in this article, but the integration of clinical risk models, current imaging techniques, and advancing biomarker research has the potential to improve the LC screening performance generally.

Reproducibility of radiomic features of pulmonary nodules between low-dose CT and conventional-dose CT

BACKGROUND

The reproducibility of radiomic features is essential to lung cancer detection. This study aimed to investigate the reproducibility of radiomic features of pulmonary nodules between low-dose computed tomography (LDCT) and conventional-dose computed tomography (CDCT).

METHODS

A total of 105 patients with 119 pulmonary nodules [39 ground-glass nodules (GGNs) and 80 solid nodules] who underwent LDCT and CDCT were retrospectively studied between September 2019 and November 2020. Pulmonary nodules were manually segmented and 1,125 radiomic features (shape, first-order intensity, texture, wavelet, and Laplacian of the Gaussian features) were extracted from both LDCT and CDCT images. The concordance correlation coefficient (CCC) was used to evaluate the reproducibility of these radiomic features.

RESULTS

Of the 1,125 radiomic features considered, 35.5% (399 of 1,125) and 41.5% (467 of 1,125) were reproducible (CCC ≥0.85) for GGNs and solid nodules, respectively. The intensity, texture, and wavelet features of solid nodules were more reproducible than those of GGNs. The mean CCC values for intensity and texture features of solid nodules were of 0.85 and above, whereas the mean values for those of GGNs were of less than 0.85. After Gaussian kernel (σ =2) preprocessing, the CCC of intensity and texture features of GGNs improved from 0.77 to 0.90, and 84.9% (79 of 93) of the radiomic features were reproducible (mean CCC increase from 0.84±0.13 to 0.92±0.08 for intensity features, and from 0.75±0.15 to 0.89±0.11 for texture features). Wavelet features had the lowest CCCs for both GGNs and solid nodules.

CONCLUSIONS

The majority of the radiomic feature classes of solid pulmonary nodules have a high level of reproducibility between LDCT and CDCT. However, LDCT should not be used as an alternative to CDCT in the radiomic study of GGNs.

Integrating Biological and Radiological Data in a Structured Repository: a Data Model Applied to the COSMOS Case Study

Integrating the information coming from biological samples with digital data, such as medical images, has gained prominence with the advent of precision medicine. Research in this field faces an ever-increasing amount of data to manage and, as a consequence, the need to structure these data in a functional and standardized fashion to promote and facilitate cooperation among institutions. Inspired by the Minimum Information About BIobank data Sharing (MIABIS), we propose an extended data model which aims to standardize data collections where both biological and digital samples are involved. In the proposed model, strong emphasis is given to the cause-effect relationships among factors as these are frequently encountered in clinical workflows. To test the data model in a realistic context, we consider the Continuous Observation of SMOking Subjects (COSMOS) dataset as case study, consisting of 10 consecutive years of lung cancer screening and follow-up on more than 5000 subjects. The structure of the COSMOS database, implemented to facilitate the process of data retrieval, is therefore presented along with a description of data that we hope to share in a public repository for lung cancer screening research.

Radiomic Phenotypes for Improving Early Prediction of Survival in Stage III Non-Small Cell Lung Cancer Adenocarcinoma after Chemoradiation

We evaluate radiomic phenotypes derived from CT scans as early predictors of overall survival (OS) after chemoradiation in stage III primary lung adenocarcinoma. We retrospectively analyzed 110 thoracic CT scans acquired between April 2012-October 2018. Patients received a median radiation dose of 66.6 Gy at 1.8 Gy/fraction delivered with proton (55.5%) and photon (44.5%) beam treatment, as well as concurrent chemotherapy (89%) with carboplatin-based (55.5%) and cisplatin-based (36.4%) doublets. A total of 56 death events were recorded. Using manual tumor segmentations, 107 radiomic features were extracted. Feature harmonization using ComBat was performed to mitigate image heterogeneity due to the presence or lack of intravenous contrast material and variability in CT scanner vendors. A binary radiomic phenotype to predict OS was derived through the unsupervised hierarchical clustering of the first principal components explaining 85% of the variance of the radiomic features. C-scores and likelihood ratio tests (LRT) were used to compare the performance of a baseline Cox model based on ECOG status and age, with a model integrating the radiomic phenotype with such clinical predictors. The model integrating the radiomic phenotype (C-score = 0.69, 95% CI = (0.62, 0.77)) significantly improved (p<0.005) upon the baseline model (C-score = 0.65, CI = (0.57, 0.73)). Our results suggest that harmonized radiomic phenotypes can significantly improve OS prediction in stage III NSCLC after chemoradiation.

Preliminary study of 3 T-MRI native T1-mapping radiomics in differential diagnosis of non-calcified solid pulmonary nodules/masses

BACKGROUND

Cumulative CT radiation damage was positively correlated with increased tumor risks. Although it has recently been known that non-radiation MRI is alternative for pulmonary imaging. There is little known about the value of MRI T1-mapping in the diagnosis of pulmonary nodules. This article aimed to investigate the value of native T1-mapping-based radiomics features in differential diagnosis of pulmonary lesions.

METHODS

73 patients underwent 3 T-MRI examination in this prospective study. The 99 pulmonary lesions on native T1-mapping images were segmented twice by one radiologist at indicated time points utilizing the in-house semi-automated software, followed by extraction of radiomics features. The inter-class correlation coefficient (ICC) was used for analyzing intra-observer's agreement. Dimensionality reduction and feature selection were performed via univariate analysis, and least absolute shrinkage and selection operator (LASSO) analysis. Then, the binary logical regression (LR), support vector machine (SVM) and decision tree classifiers with the input of optimal features were selected for differentiating malignant from benign lesions. The receiver operative characteristics (ROC) curve, area under the curve (AUC), sensitivity, specificity and accuracy were calculated. Z-test was used to compare differences among AUCs.

RESULTS

107 features were obtained, of them, 19.5% (n = 21) had relatively good reliability (ICC ≥ 0.6). The remained 5 features (3 GLCM, 1 GLSZM and 1 shape features) by dimensionality reduction were useful. The AUC of LR was 0.82(95%CI: 0.67-0.98), with sensitivity, specificity and accuracy of 70%, 85% and 80%. The AUC of SVM was 0.82(95%CI: 0.67-0.98), with sensitivity, specificity and accuracy of 70, 85 and 80%. The AUC of decision tree was 0.69(95%CI: 0.49-0.87), with sensitivity, specificity and accuracy of 50, 85 and 73.3%.

CONCLUSIONS

The LR and SVM models using native T1-mapping-based radiomics features can differentiate pulmonary malignant from benign lesions, especially for uncertain nodules requiring long-term follow-ups.

A segmentation tool for pulmonary nodules in lung cancer screening: Testing and clinical usage

PURPOSE

With the future goal of defining a large dataset based on low-dose CT with labelled pulmonary lesions for lung cancer screening (LCS) research, the aim of this work is to propose and evaluate into a clinical context a tool for semi-automatic segmentation able to facilitate the process of labels collection from a LCS study (COSMOS, Continuous Observation of SMOking Subjects).

METHODS

Considering a preliminary set of manual annotations, a segmentation model based on a 2D-Unet was trained from scratch. Contour quality of the final 2D-Unet was assessed on an internal test set of manual annotations and on a subset of the public available LIDC dataset used as external test set. The tool for semi-automatic segmentation was then designed integrating the tested model into a Graphical User Interface. According to the opinion of two clinical users, the percentage of lesions properly contoured through the tool was quantified (Acceptance Rate, AR). The variability between segmentations derived by the two readers was estimated as mean percentage of difference (MPD) between the two sets of volumes and comparing the likelihood of malignancy derived from Volume Doubling Time (VDT).

RESULTS

Performance in test sets were found similar (DICE ~ 0.75(0.15)). Accordingly, a good mean AR (80.1%) resulted from the two readers. Variability in terms of MPD was equal to 23.6% while 2.7% was the VDTs percentage of disagreement.

CONCLUSIONS

A semi-automatic segmentation tool was developed and its applicability evaluated into a clinical context demonstrating the efficacy of the tool in facilitating the collection of labelled data.

A Low-Dose CT-Based Radiomic Model to Improve Characterization and Screening Recall Intervals of Indeterminate Prevalent Pulmonary Nodules

Lung cancer (LC) is currently one of the main causes of cancer-related deaths worldwide. Low-dose computed tomography (LDCT) of the chest has been proven effective in secondary prevention (i.e., early detection) of LC by several trials. In this work, we investigated the potential impact of radiomics on indeterminate prevalent pulmonary nodule (PN) characterization and risk stratification in subjects undergoing LDCT-based LC screening. As a proof-of-concept for radiomic analyses, the first aim of our study was to assess whether indeterminate PNs could be automatically classified by an LDCT radiomic classifier as solid or sub-solid (first-level classification), and in particular for sub-solid lesions, as non-solid versus part-solid (second-level classification). The second aim of the study was to assess whether an LCDT radiomic classifier could automatically predict PN risk of malignancy, and thus optimize LDCT recall timing in screening programs. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, positive predictive value, negative predictive value, sensitivity, and specificity. The experimental results showed that an LDCT radiomic machine learning classifier can achieve excellent performance for characterization of screen-detected PNs (mean AUC of 0.89 ± 0.02 and 0.80 ± 0.18 on the blinded test dataset for the first-level and second-level classifiers, respectively), providing quantitative information to support clinical management. Our study showed that a radiomic classifier could be used to optimize LDCT recall for indeterminate PNs. According to the performance of such a classifier on the blinded test dataset, within the first 6 months, 46% of the malignant PNs and 38% of the benign ones were identified, improving early detection of LC by doubling the current detection rate of malignant nodules from 23% to 46% at a low cost of false positives. In conclusion, we showed the high potential of LDCT-based radiomics for improving the characterization and optimizing screening recall intervals of indeterminate PNs.

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 review in radiomics: Making personalized medicine a reality via routine imaging

Radiomics is the quantitative analysis of standard-of-care medical imaging; the information obtained can be applied within clinical decision support systems to create diagnostic, prognostic, and/or predictive models. Radiomics analysis can be performed by extracting hand-crafted radiomics features or via deep learning algorithms. Radiomics has evolved tremendously in the last decade, becoming a bridge between imaging and precision medicine. Radiomics exploits sophisticated image analysis tools coupled with statistical elaboration to extract the wealth of information hidden inside medical images, such as computed tomography (CT), magnetic resonance (MR), and/or Positron emission tomography (PET) scans, routinely performed in the everyday clinical practice. Many efforts have been devoted in recent years to the standardization and validation of radiomics approaches, to demonstrate their usefulness and robustness beyond any reasonable doubts. However, the booming of publications and commercial applications of radiomics approaches warrant caution and proper understanding of all the factors involved to avoid "scientific pollution" and overly enthusiastic claims by researchers and clinicians alike. For these reasons the present review aims to be a guidebook of sorts, describing the process of radiomics, its pitfalls, challenges, and opportunities, along with its ability to improve clinical decision-making, from oncology and respiratory medicine to pharmacological and genotyping studies.

Structural and functional radiomics for lung cancer

INTRODUCTION

Lung cancer ranks second in new cancer cases and first in cancer-related deaths worldwide. Precision medicine is working on altering treatment approaches and improving outcomes in this patient population. Radiological images are a powerful non-invasive tool in the screening and diagnosis of early-stage lung cancer, treatment strategy support, prognosis assessment, and follow-up for advanced-stage lung cancer. Recently, radiological features have evolved from solely semantic to include (handcrafted and deep) radiomic features. Radiomics entails the extraction and analysis of quantitative features from medical images using mathematical and machine learning methods to explore possible ties with biology and clinical outcomes.

METHODS

Here, we outline the latest applications of both structural and functional radiomics in detection, diagnosis, and prediction of pathology, gene mutation, treatment strategy, follow-up, treatment response evaluation, and prognosis in the field of lung cancer.

CONCLUSION

The major drawbacks of radiomics are the lack of large datasets with high-quality data, standardization of methodology, the black-box nature of deep learning, and reproducibility. The prerequisite for the clinical implementation of radiomics is that these limitations are addressed. Future directions include a safer and more efficient model-training mode, merge multi-modality images, and combined multi-discipline or multi-omics to form "Medomics."

Molecular biomarkers in early stage lung cancer

Low dose computed tomography (LDCT) screening, together with the recent advances in targeted and immunotherapies, have shown to improve non-small cell lung cancer (NSCLC) survival. Furthermore, screening has increased the number of early stage-detected tumors, allowing for surgical resection and multimodality treatments when needed. The need for improved sensitivity and specificity of NSCLC screening has led to increased interest in combining clinical and radiological data with molecular data. The development of biomarkers is poised to refine inclusion criteria for LDCT screening programs. Biomarkers may also be useful to better characterize the risk of indeterminate nodules found in the course of screening or to refine prognosis and help in the management of screening detected tumors. The clinical implications of these biomarkers are still being investigated and whether or not biomarkers will be included in further decision-making algorithms in the context of screening and early lung cancer management still needs to be determined. However, it seems clear that there is much room for improvement even in early stage lung cancer disease-free survival (DFS) rates; thus, biomarkers may be the key to refine risk-stratification and treatment of these patients. Clinicians’ capacity to register, integrate, and analyze all the available data in both high risk individuals and early stage NSCLC patients will lead to a better understanding of the disease’s mechanisms, and will have a direct impact in diagnosis, treatment, and follow up of these patients. In this review, we aim to summarize all the available data regarding the role of biomarkers in LDCT screening and early stage NSCLC from a multidisciplinary perspective. We have highlighted clinical implications, the need to combine risk stratification, clinical data, radiomics, molecular information and artificial intelligence in order to improve clinical decision-making, especially regarding early diagnostics and adjuvant therapy. We also discuss current and future perspectives for biomarker implementation in routine clinical practice.

Radiomics and Dosiomics for Predicting Local Control after Carbon-Ion Radiotherapy in Skull-Base Chordoma

Skull-base chordoma (SBC) can be treated with carbon ion radiotherapy (CIRT) to improve local control (LC). The study aimed to explore the role of multi-parametric radiomic, dosiomic and clinical features as prognostic factors for LC in SBC patients undergoing CIRT. Before CIRT, 57 patients underwent MR and CT imaging, from which tumour contours and dose maps were obtained. MRI and CT-based radiomic, and dosiomic features were selected and fed to two survival models, singularly or by combining them with clinical factors. Adverse LC was given by in-field recurrence or tumour progression. The dataset was split in development and test sets and the models' performance evaluated using the concordance index (C-index). Patients were then assigned a low- or high-risk score. Survival curves were estimated, and risk groups compared through log-rank tests (after Bonferroni correction α = 0.0083). The best performing models were built on features describing tumour shape and dosiomic heterogeneity (median/interquartile range validation C-index: 0.80/024 and 0.79/0.26), followed by combined (0.73/0.30 and 0.75/0.27) and CT-based models (0.77/0.24 and 0.64/0.28). Dosiomic and combined models could consistently stratify patients in two significantly different groups. Dosiomic and multi-parametric radiomic features showed to be promising prognostic factors for LC in SBC treated with CIRT.