Predictive models and biomarkers for early stage and oligometastatic non-small cell lung cancer patients treated with stereotactic body radiation therapy, using machine learning

Stereotactic body radiation therapy (SBRT) is currently the alternative for inoperable early-stage and oligometastatic non-small cell lung cancer (NSCLC) patients. Unfortunately, a minority of patients are not good responders. Even though physicians use clinical variables and semantic radiological features to make treatment decisions, medical images contain a wealth of personalized pathophysiological information that can be extracted and used for clinical decision support systems. Indeed, radiomic features can be used to generate predictive algorithms and biomarkers that can determine treatment outcomes and stratify patients to their therapeutic options. Additionally, while some features have shown predictive potential for SBRT-treated NSCLC tumors, radiomics features are not always stable. Thus, validating predictive biomarkers for applications in our institution is important. This study investigated the radiomic features of images and the clinical parameters obtained from early-stage and oligometastatic NSCLC patients who underwent SBRT, to predict different response outcomes. A single-institution retrospective review of patients’ medical records (n = 98 patients; median age = 76 years; male/female ratio = 46/52; 116 lesions treated with SBRT from 2009 to 2022) was conducted. The radiomics features (107 features) extracted from CT planning scans with PyRadiomics, along with the patients’ clinical data were collected. The response to SBRT was analyzed from follow-up scans. The local response was defined as per RECIST criteria. The regional progression was defined as the appearance of new tumors and the worsening of the disease across other regions of the lung. Distant progression was defined as the spreading of disease outside the lung. The adaptive synthetic (ADASYN) sampling method corrected the imbalance in responses. Classification models, which included support vector machine (SVM) with linear or radial basis function (RBF) kernels, random forest, adaptive boosting (AdaBoost) and multi-layer perceptron (MLP), were used. Models were trained using a 5-fold cross-validation scheme. Their performances were measured with the areas under the curve (AUC) of receiver operating characteristic (ROC) plots on the validation folds. Using permutation feature importance, predictive biomarkers were identified.Highly performing models were generated for the prediction of local response and distant progression; respectively, the best models had AUCs of 0.94+/-0.05 and 0.98+/-0.02. When oligometastatic patients were omitted, the best models for local response (AUC: 0.95+/-0.06) and distant progression (AUC: 0.99+/-0.0) were as predictive. For local response models, the treatment site and the performance status, along with radiomic features such as first-order root-mean-squared-intensity, first-order skewness and GLSZM gray-level-non-uniformity, emerged as predictive. For distant progression models, clinical variables including the treatment site, the initial staging and the performance status were predictive. The predictive models created, and the biomarkers identified could be used in clinical support decision systems. Indeed, these tools could spare the minority of patients who do not benefit from SBRT, thus helping physicians to adjust their patients’ treatment course. Furthermore, consistent with previous research, root-mean-squared-intensity and skewness were found to be predictive radiomic biomarkers. By validating these features with our cohort, we showed the features’ ability to maintain their predictive capabilities in an external setting

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.