Wencheng Shao

Background: Accurate histological subtyping of esophageal cancer is critical for treatment selection and prognostication. Whole-slide images (WSIs) provide rich morphological information, but they are extremely large and lack region-level annotations. We evaluated an interpretable deep learning pipeline for classifying esophageal squamous cell carcinoma (ESCC) and esophageal adenocarcinoma (EAC) using WSIs. Methods: This retrospective study included 139 hematoxylin-and-eosin-stained WSIs from 139 patients (58 with EAC and 81 with ESCC). To differentiate EAC from ESCC, tissue tiles were generated and fixed-length embeddings were extracted using the original frozen Google ViT. During model fitting, each slide was represented by a randomly sampled bag of 128 tile embeddings and slide-level prediction was performed using an attention-gated multiple-instance learning model. We employed focal loss and weighted sampling strategies and evaluated internal performance using stratified fivefold cross-validation. The five original MIL checkpoints were applied to a balanced 50-case non-TCGA public cohort with an internal-only Platt mapping and post hoc full-cohort threshold optimization. Results: The mean AUC was high in the training sets (0.98 ± 0.01) and internal validation folds (0.95 ± 0.05), with the best validation fold exhibiting an AUC of 0.99 and an accuracy of 96.4%. Neither normalized attention entropy ( p = 0.364) nor maximum attention ( p = 0.423) differed significantly between histologies. In post hoc full-cohort threshold optimization, internally calibrated external scores achieved apparent 80.0% accuracy and balanced accuracy, with adenocarcinoma sensitivity of 76.0% and ESCC sensitivity of 84.0% . Conclusions: This model demonstrated potential as an interpretable slide-level screening approach for ESCC versus EAC classification, but the external operating threshold was domain dependent. Prospective multicenter validation using a prespecified cutoff is required.

Objective: This study aimed to investigate whether baseline skeletal muscle status and malnutrition in patients with locally advanced head and neck cancer (HNC) are associated with short‐term treatment outcomes and acute radiation oral mucositis (OM). Methods: We retrospectively analyzed data from 68 patients with stage III–IV HNC treated at a single center with definitive radiotherapy (66–70 Gy) with or without chemotherapy (induction and/or concurrent) between July 2021 and July 2023. Malnutrition was diagnosed using the Global Leadership Initiative on Malnutrition (GLIM) criteria and Patient‐Generated Subjective Global Assessment (PG‐SGA). Short‐term response was assessed at first post‐treatment imaging after radiotherapy according to RECIST 1.1, and OM severity was graded using World Health Organization criteria. Univariate and multivariate logistic regression analyses evaluated the correlation between short‐term response and grades III–IV OM. Results: Of the 68 patients, 32.4% had a lower skeletal muscle index (SMI), 26.5% had lower hand grip strength (HGS), 20.6% (GLIM) and 58.8% (PG‐SGA) had malnutrition. At the end of treatment, 64.7% of the patients achieved partial remission, and 35.3% achieved a stable condition. Low HGS was associated with low short‐term efficacy ( OR = 0.31, 95% CI = 0.10–0.95, p = 0.040); however, this difference did not reach statistical significance in the multivariate model. A total of 41.2% of patients had Grade III–IV OM; however, OM was not significantly associated with low SMI, HGS, or malnutrition. Conclusion: Baseline skeletal muscle status and malnutrition did not consistently predict short‐term treatment outcomes or severe radiation‐induced OM in patients with locally advanced HNC. However, these results suggest a potential relationship between low HGS and treatment response. Further research with larger sample sizes and longitudinal evaluations is needed to clarify muscle and nutritional status impact on multidimensional outcomes in this patient population.

Background: Computed Tomography (CT) imaging is essential for disease detection but carries a risk of cancer due to X‐ray exposure. Typically, assessing this risk requires segmentation of the internal organ contours to predict organ doses, which hinders its clinical application. This study introduces a method that uses support vector regression (SVR) models trained on skin outline radiomic features to predict organ doses without organ segmentation, thus streamlining the process for clinical use.Methods: CT scans of the head and abdomen were used to extract radiomic features of the skin outline. These features were used as inputs, with organ doses from Monte Carlo simulations as benchmarks to train the SVR models for predicting organ doses. The accuracy of the models was evaluated using the mean absolute percentage error (MAPE) and coefficient of determination (R²).Results: The results showed a high precision in dose prediction for various organs, including the brain (MAPE: 1.5%, R²: 0.9), eyes (MAPE: 5%, R²: 0.84), lens (MAPE: 5%, R²: 0.82), bowel (MAPE: 6%, R²: 0.84), kidneys (MAPE: 7.5%, R²: 0.7), and liver (MAPE: 8%, R²: 0.67). Internal organ disturbances had a minimal impact on accuracy.Conclusions: The SVR models efficiently predicted patient‐specific organ doses from CT scans, offering a user‐friendly tool for rapid segmentation‐free dose prediction. This innovation can significantly enhance clinical efficiency and accessibility in predicting patient‐specific organ doses using CT.

Background: CT examinations are commonly utilized for the diagnosis of internal diseases. The X-rays emitted during CT scans can elevate the risks of developing solid cancers by causing DNA damage. The risk of CT scan-induced solid cancers is intricately linked to the organ doses specific to each patient. The Support Vector Regression (SVR) algorithm exhibits the capability to swiftly and accurately predict organ doses. Kernel functions, including linear, polynomial, and radial basis (RBF) functions, play a crucial role in the overall performance of SVR when predicting patient-specific organ doses from CT scans. Therefore, it is imperative to investigate the influence of kernel selection on the comprehensive predictive effectiveness of SVR. Purpose: This study investigates the impact of kernel functions on the predictive performance of SVR models trained by radiomics features, and to pinpoint the optimal kernel function for predicting patient-specific organ doses from CT scans. Methods: CT images from head and abdominal CT scans were processed using DeepViewer^, an auto-segmentation tool for defining regions of interest (ROIs) within their organs. Radiomics features were extracted from the CT data and ROIs. Benchmark organ doses were calculated through Monte Carlo simulations. SVR models, utilizing the radiomics features, were trained with linear-, polynomial-, and RBF kernels to predict patient-specific organ doses from CT scans. The robustness of the SVR prediction was examined by applying 25 random sample splits with each kernel. The mean absolute percentage error (MAPE) and coefficient of determination (R²) were compared among the kernels to identify the optimal kernel. Results: The linear kernel obtains better overall predictive performance than the polynomial and RBF kernels. The SVR trained with the linear kernel function achieves lower MAPE values, below 5% for head organs and under 6.8% for abdominal organs. Furthermore, it shows higher R² values exceeding 0.85 for head organs and going beyond 0.8 for abdominal organs. Conclusions: Kernel selection severely impact the overall performance of SVR models. The optimal kernel varies with CT scanned parts and organ types indicating the necessity to conduct organ-specific kernel selection.

This study aims to establish and validate a cross‐institutional prediction system for estimating patient‐specific organ doses from chest CT scans. By collaborating across multiple institutions, the study seeks to develop models that allow rapid online estimation of organ doses, eliminating the need for complex organ segmentation. Researchers delineate skin outlines from chest CT images obtained from two different institutions. Radiomics features are extracted from the chest CT data and skin contours. Organ doses are computed using Monte Carlo simulations as reference organ doses. Single‐ and cross‐institutional support vector regression (SVR) models are trained with radiomics features to predict organ doses from chest CT scans. Model performance is assessed using metrics like maean absolute percentage error (MAPE) and R‐squared (R²). For chest organs (lungs, heart, spinal cord, trachea, and esophagus), single‐institutional SVR models achieve MAPE values ranging from 4.72% to 15.31% and R² values from 0.73 to 0.93. Cross‐institutional models have MAPE values in the range of 2.16–7.49% and R² values from 0.84 to 0.99. SVR models trained with radiomics features from skin outlines can reliably estimate organ doses from chest CT scans. The proposed method is applicable across different institutions, and cross‐institutional models enhance predictive accuracy, generality, and robustness.

Objectives . This study aims to develop a method for predicting patient-specific head organ doses by training a support vector regression (SVR) model based on radiomics features and graphics processing unit (GPU)-calculated reference doses. Methods . In this study, 237 patients who underwent brain CT scans were selected, and their CT data were transferred to an autosegmentation software to segment head regions of interest (ROIs). Subsequently, radiomics features were extracted from the CT data and ROIs, and the benchmark organ doses were computed using fast GPU-accelerated Monte Carlo (MC) simulations. The SVR organ dose prediction model was then trained using the radiomics features and benchmark doses. For the predicted organ doses, the relative root mean squared error (RRMSE), mean absolute percentage error (MAPE), and coefficient of determination ( R ² ) were evaluated. The robustness of organ dose prediction was verified by changing the patient samples on the training and test sets randomly. Results . For all head organs, the maximal difference between the reference and predicted dose was less than 1 mGy. For the brain, the organ dose was predicted with an absolute error of 1.3%, and the R ² reached up to 0.88. For the eyes and lens, the organ doses predicted by SVR achieved an RRMSE of less than 13%, the MAPE ranged from 4.5% to 5.5%, and the R ² values were more than 0.7. Conclusions . Patient-specific head organ doses from CT examinations can be predicted within one second with high accuracy, speed, and robustness by training an SVR using radiomics features.

Orbital meningioma is a rare type of orbital tumor with high invasiveness and recurrence rates, making it extremely challenging to treat. Due to the special location of the disease, surgery often cannot completely remove the tumor, requiring postoperative radiation therapy. Here, we report a case of an elderly male patient with right-sided proptosis, visual impairment, and diplopia. Imaging diagnosis revealed a space-occupying lesion in the extraconal space of the right orbit. Pathological and immunohistochemical examination of the resected tumor confirmed it as a grade 3 anaplastic meningioma. Two months after surgery, the patient complained of right eye swelling and a magnetic resonance imaging (MRI) scan showed a recurrence of the tumor. The patient received helical tomotherapy (TOMO) in the postoperative tumor bed and high-risk areas within the orbit with a total dose of 48Gy. However, there was no significant improvement in the patient’s right eye swelling, and the size of the recurrent lesion showed no significant change on imaging. Gamma knife multifractionated stereotactic radiosurgery (MF-SRS) was then given to the recurrent lesion with 50% prescription dose 13.5Gy/3f, once every other day. An imaging diagnosis performed 45 days later showed that the tumor had disappeared completely. The patient’s vision remained unchanged, but diplopia was significantly relieved after MF-SRS. We propose a new hybrid treatment model for recurrent orbital meningioma, where conventional radiation therapy ensures local control of high-risk areas around the postoperative cavity, and MF-SRS maximizes the radiation dose to recurrent lesion areas while protecting surrounding tissues and organs.

This study aims to develop a method for verifying site-specific and/or beam path specific proton beam range, which could reduce range uncertainty margins and the associated treatment complications. It investigates the range uncertainties from both CT HU to relative stopping power conversion and patient positioning errors for prostate treatment using pelvic-like biological phantoms. Three 25 × 14 × 12 cm ³ phantoms, made of fresh animal tissues mimicking the pelvic anatomies of prostate patients, were scanned with a general electric CT simulator. A 22 cm circular passive scattering beam with 29 cm range and 8 cm modulation width was used to measure the water equivalent path lengths (WEPL) through the phantoms at multiple points using the dose extinction method with a MatriXXPT detector. The measured WEPLs were compared to those predicted by TOPAS simulations and ray-tracing WEPL calculations. For the three phantoms, the WEPL differences between measured and theoretical prediction (WDMT) are below 1.8% for TOPAS, and 2.5% for ray-tracing. WDMT varies with phantom anatomies by about 0.5% for both TOPAS and ray-tracing. WDMT also correlates with the tissue types of a specific treated region. For the regions where the proton beam path is parallel to sharp bone edges, the WDMTs of TOPAS and ray-tracing respectively reach up to 1.8% and 2.5%. For the region where proton beams pass through just soft tissues, the WDMT is mostly less than 1% for both TOPAS and ray-tracing. For prostate treatments, range uncertainty depends on the tissue types within a specific treated region, patient anatomies and the range calculation methods in the planning algorithms. Our study indicates range uncertainty is less than 2.5% for the whole treated region with both ray-tracing and TOPAS, which suggests the potential to reduce the current 3.5% range uncertainty margin used in the clinics by at least 1% even for single-energy CT data.