This paper presents a new technique for improving the machine learning (ML)-based cardiac disease classification model by optimizing the weights of cardiac-affecting parameters (CAPs), such as systolic and diastolic blood pressure, cholesterol, BMI, SpO2, and other clinical and physiological parameters. Now, it is well established that accurate and robust prediction of cardiac diseases requires analysis of electrocardiograms along with other cardiac-affecting parameters. A systematic study is still required to determine the weightage of these parameters in the onset of cardiac disease. This paper introduces a scientific technique to determine and optimize CAPs weight; the results help develop a robust, accurate algorithm for early detection of cardiovascular disease. CAPs weight optimization is achieved using a hybrid metaheuristic (LSTM + PSO) technique. This technique significantly contributes to the development of an improved method for early detection of cardiac diseases. The improvement is achieved by determining the weights of CAPs and applying these weights for fine-tuning the ML-based CatBoost, XGBoost, TapNet, and SVM classification models. The study integrates the perspectives of 200 cardiologists on 13 CAPs. The variability in doctors’ opinions creates a challenge in determining the best weights for each CAP, and it is resolved with the optimized CAPs technique. It is found that the optimized CatBoost classifier algorithm achieves the highest accuracy of 99.8
In vivo dosimetry (IVD) is essential for verifying treatment accuracy and current methods only identify discrepancies after delivery. Predicting IVD results beforehand offers proactive risk anticipation. This study aimed to develop a machine learning (ML) model for predicting electronic-portal-imaging-device (EPID)-based IVD results. Input predictors included volumetric modulated arc therapy (VMAT) plan complexity metrics, gamma passing rate (GPRs) from EPID-based patient-specific quality assurance (PSQA), and daily machine output quality assurance (QA) data for each treatment fraction. Seventeen VMAT plan complexity metrics were calculated from DICOM-RT files, and GPRs were obtained from forty head and neck cancer patients. The IVD results, a set of point-dose deviations between reconstructed and planned dose distributions, served as the input response. An ensemble-bagged-tree regression model was used to develop the prediction model. Model performance was evaluated using root mean square error (RMSE), mean absolute error (MAE), and R-squared. Spearman’s rank correlation assessed relationships between predictors and responses. In the test set, the model achieved RMSE of 2.9
CT simulators offer an extended field-of-view (eFOV) to include immobilization devices and anatomy beyond the scan FOV (sFOV). This study evaluated its impact on image quality, stopping power ratio (SPR), and proton dose calculations. CT images were reconstructed into the sFOV and three eFOVs. Image quality was assessed using the CATPHAN phantom. Hounsfield unit (HU)-to-SPR curves were generated using the stoichiometric method. HU and SPR variations were evaluated using homogeneous and heterogeneous phantoms extending beyond the sFOV. Proton dose calculations using 50-cm sFOV and 80-cm eFOV images were compared in breast and lung cases using dose differences and three-dimensional gamma analysis. Geometric distortion remained below 0.5 mm at the center but increased to 1.7 mm near the FOV edge. Spatial resolution, uniformity, and contrast were comparable across FOVs. HU-to-SPR calibration curves differed by ≤0.6
Cardiac murmur detection and severity grading are critical for early cardiovascular screening; however, existing studies often address these tasks separately and lack sufficient post-hoc explainability. This study aimed to develop a deep learning framework with post-hoc explainability for joint murmur detection and severity grading from phonocardiogram recordings. Heart sound recordings from the CirCor DigiScope dataset were preprocessed using Butterworth band-pass filtering, normalization, and sliding-window segmentation, and transformed into CWT-based time–frequency representations. An improved Swin Transformer, termed ST-CAG (Swin Transformer with Cross-Task Attention and Gating), was proposed for simultaneous murmur detection and three-class severity grading. The framework integrates a spatial attention module for enhancing murmur-related features, a Cross-Task Attention module for information exchange between tasks, and a Grade Gate mechanism for improving severity prediction consistency. Post-hoc explainability was investigated using spatial attention heatmap visualization. ST-CAG achieved accuracies of 98.99
Enhanced Leaf Modelling (ELM), introduced in Varian Eclipse v18, provides a refined representation of MLC transmission by explicitly modelling physical leaf geometry, interleaf leakage, and leaf-tip effects for Halcyon, replacing the traditional dosimetric leaf gap approximation. Unlike C-arm systems, Halcyon employs no collimator jaws, making out-of-field dose more exclusively dependent on MLC transmission characteristics and increasing the relative importance of accurate MLC modelling. Local commissioning supported the implementation of ELM on Halcyon and demonstrated improved agreement between calculated and measured dose distributions in the commissioning tests. These commissioning findings were separate form the present analysis, which characterises the dosimetric impact of ELM in clinically representative multi-metastatic stereotactic radiotherapy (MMSRT) plans rather than independently validating their physical accuracy. Ten MMSRT patients (2–6 lesions) were replanned using two calculation models: the standard Acuros XBv16 and v18 (with ELM). For each, two optimisation strategies were compared, manual and automatic normal tissue optimisation (NTO), to assess robustness across planning methods. The low-dose valley, defined as [Brain-PTVs], represented normal brain between targets. Dose-volume metrics were extracted across V50
