Accurate pre-treatment dose prediction is essential for efficient radiotherapy planning. Although deep learning models have advanced automated dose distribution, comprehensive multi-tumor analyses remain scarce. This study assesses deep learning models for dose prediction across diverse tumor types, combining objective and subjective evaluation methods. We included 622 patients with planning data across various tumor sites: nasopharyngeal carcinoma (n = 29), esophageal carcinoma (n = 82), left-sided breast carcinoma (n = 107), right-sided breast carcinoma (n = 95), cervical carcinoma treated with radical radiotherapy (n = 84), postoperative cervical carcinoma (n = 122), and rectal carcinoma (n = 103). Dose predictions were generated using U-Net, Flex-Net, and Highres-Net models, with data split into training (60%), validation (20%), and testing (20%) sets. Quantitative comparisons used normalized dose difference (NDD) and dose-volume histogram (DVH) metrics, and qualitative assessments by radiation oncologists were performed on the testing set. Predicted and clinical doses correlated well, with NDD values under 3% for tumor targets in nasopharyngeal, breast, and postoperative cervical cancer. Qualitative assessments revealed that U-Net, Flex-Net, and Highres-Net achieved the highest accuracy in cervical radical, breast/rectal/postoperative cervical, and nasopharyngeal/esophageal cancers, respectively. Among the test cases (n = 123), 53.7% were deemed clinically acceptable and 32.5% required minor adjustments. The “Best Selection” approach, combining strengths of all three models, raised clinical acceptance to 62.6%. This study demonstrates that automated dose prediction can provide a robust starting point for rapid plan generation. Leveraging model-specific strengths through the “Best Selection” approach enhances prediction accuracy and shows potential to improve clinical efficiency across multiple tumor types.

Radiation therapy can be categorised by particle type into photon, proton and heavy ion therapies. Proton radiotherapy is highlighted due to its unique physical properties, such as the Bragg peak and minimal exit dose, which offer superior dose distribution. This makes proton radiotherapy especially advantageous for treating tumours near vital organs with complex structures, such as gliomas near the brain, nasopharyngeal carcinoma near the brainstem and mediastinal tumours near the heart. Proton irradiation can induce distant effects through immunogenicity within the target area. The reduced low‐dose zone outside the target provides better lymphatic system protection and immune benefits. Additionally, combining proton radiotherapy with immunotherapy may offer further biological advantages. These features make proton radiotherapy a promising option in cancer treatment. This article may aid in the understanding of proton radiotherapy and its immune effects and lead to new effective options for tumour treatment.

The aim of this study was to determine whether functional dose-volume histograms (FDVHs) are valuable for predicting radiation pneumonitis (RP), and to identify whether FDVHs have advantages over conventional dose-volume histograms (DVHs) for the prediction of RP in patients with locally advanced non-small cell lung cancer (LANSCLC). Fifty-seven patients with LANSCLC undergoing functional image-guided late-course accelerated hyperfractionated radiotherapy were enrolled. The grade of RP was evaluated according to the Common Toxicity Criteria 3.0. To identify predictive factors of RP, the FDVHs, including the volume of the functional lung receiving 5 Gy (FV5) through 50 Gy (FV50), mean perfusion-weighted lung dose (MPWLD) and functional normal tissue complication probability (FNTCP), were analyzed and compared to their counterparts [total lung receiving 5 Gy (V5) through 50 Gy (V50), mean lung dose (MLD) and normal tissue complication probability (NTCP)] derived from conventional DVHs. Univariate analysis revealed that V5-V40, MLD, NTCP and FV5-FV50, MPWLD, FNTCP were all statistically significant relative to the development of RP (all p<0.05). Multivariate analysis showed that only MLD and FV15 were associated with RP (p=0.001 and 0.044, respectively). Receiver operator characteristic curve anaysis indicated that almost all of the FDVHs had larger areas under the curve compared to the DVHs, although no statistically significant difference was observed (p-value ranged from 0.066 to 0.951). FDVHs are valuable for predicting RP with the predictive efficiency equivalent to or slightly advantageous over conventional DVHs. More homogeneous studies involving larger numbers of patients are required to further assess the value of FDVHs for predicting RP.

Purpose: To assess the out‐of‐field surface and internal dose of the 1.5 T MR‐Linac compared to the conventional external beam linac using optically stimulated luminescence dosimeters (OSLDs), and evaluate the out‐of‐field dose calculation accuracy of the Monaco treatment planning system (TPS) of the 1.5T MR‐Linac.Methods: A cubic solid water phantom, with OSLDs on the surface, was vertically irradiated by MR‐Linac square fields with different sizes. In addition, OSLDs were arranged out of the beam edges in four directions. An anthropomorphic adult phantom, with 125 cm³ simulated volume, was irradiated in four orthogonal directions by both MR‐Linac and conventional linac at the head, thoracic, and pelvic sites. Out‐of‐field doses were measured by OSLDs on both the surface and internal emulational organs at risk (OARs). The results were compared to the simulated dose from Monaco TPS.Results: At different field sizes (5 × 5 to 20 × 20 cm²) and distances (1 to 10 cm) to beam edge, the out‐of‐field surface dose measured on MR‐Linac varied from 0.16 % (10 cm to 5 × 5 cm² edge) to 7.02 % (1 cm to 20 × 20 cm² edge) of the maximum dose laterally and from 0.14 % (10 cm to 5 × 5 cm² edge) to 8.56 % (1 cm to 20 × 20 cm² edge) of the maximum dose longitudinally. Compared to the OSLDs measured data, the Monaco TPS presented an overestimate of the out‐of‐field dose of OARs at 0–2 % isodose area on both surface and internal check points, and the overestimation gets greater as the distance increases. The underestimation was found to be 0–35% at 2–5% isodose area on both surface and internal check points. Compared to the conventional linac, MR‐Linac delivered higher average values of out‐of‐field dose on surface check points (20%, 19%, 21%) and internal simulated OARs (42%, 37%, 9%) of the anthropomorphic phantom at head, thoracic, and pelvic irradiations, respectively.Conclusions: Compared to the conventional linac, MR‐Linac has the same out‐of‐field dose distribution. However, considering the absolute dose values, MR‐Linac delivered relatively higher out‐of‐field doses on both surface and internal OARs. Additional radiation shielding to patients undergoing MR‐Linac may provide protection from out‐of‐field exposure.

Background: Ki‐67 is a key marker of tumor proliferation. This study aimed to develop machine learning models using single‐ and multi‐parameter MRI radiomic features for the preoperative prediction of Ki‐67 expression in primary central nervous system lymphoma (PCNSL), aiding prognosis and individualized treatment planning.Methods: A retrospective analysis of 74 patients was conducted using MRI scans, including T1, contrast‐enhanced T1, T2, T2‐FLAIR, DWI, and ADC sequences. Patients were categorized into high‐expression (Ki‐67 > 70%) and low‐expression (Ki‐67 ≤ 70%) groups. Tumor volumes of interest (VOIs) were manually delineated by radiologists, and 851 radiomic features were extracted using 3DSlicer. After preprocessing, including bias field correction and normalization, feature selection was performed using SelectKBest and ANOVA. Eight machine learning classifiers, including Logistic Regression, Random Forest, and SVM, were applied to single‐ and multi‐parameter datasets.Results: Multiparameter models, particularly Naive Bayes and Logistic Regression, demonstrated superior predictive performance (AUC: 0.78, 0.73; AP: 0.90, 0.83) compared to single‐parameter models. Decision curve analysis highlighted that Logistic Regression provides the highest net benefit, followed by Naive Bayes.Conclusion: Multiparameter MRI models are more accurate and stable for predicting Ki‐67 expression in PCNSL, supporting clinical decision‐making.