Predicting treatment response in multicenter non-small cell lung cancer patients based on federated learning.

BACKGROUND: Multicenter non-small cell lung cancer (NSCLC) patient data is information-rich. However, its direct integration becomes exceptionally challenging due to constraints involving different healthcare organizations and regulations. Traditional centralized machine learning methods require centralizing these sensitive medical data for training, posing risks of patient privacy leakage and data security issues. In this context, federated learning (FL) has attracted much attention as a distributed machine learning framework. It effectively addresses this contradiction by preserving data locally, conducting local model training, and aggregating model parameters. This approach enables the utilization of multicenter data with maximum benefit while ensuring privacy safeguards. Based on pre-radiotherapy planning target volume images of NSCLC patients, a multicenter treatment response prediction model is designed by FL for predicting the probability of remission of NSCLC patients. This approach ensures medical data privacy, high prediction accuracy and computing efficiency, offering valuable insights for clinical decision-making. METHODS: We retrospectively collected CT images from 245 NSCLC patients undergoing chemotherapy and radiotherapy (CRT) in four Chinese hospitals. In a simulation environment, we compared the performance of the centralized deep learning (DL) model with that of the FL model using data from two sites. Additionally, due to the unavailability of data from one hospital, we established a real-world FL model using data from three sites. Assessments were conducted using measures such as accuracy, receiver operating characteristic curve, and confusion matrices. RESULTS: The model's prediction performance obtained using FL methods outperforms that of traditional centralized learning methods. In the comparative experiment, the DL model achieves an AUC of 0.718/0.695, while the FL model demonstrates an AUC of 0.725/0.689, with real-world FL model achieving an AUC of 0.698/0.672. CONCLUSIONS: We demonstrate that the performance of a FL predictive model, developed by combining convolutional neural networks (CNNs) with data from multiple medical centers, is comparable to that of a traditional DL model obtained through centralized training. It can efficiently predict CRT treatment response in NSCLC patients while preserving privacy.

Multi-phase contrast-enhanced magnetic resonance image-based radiomics-combined machine learning reveals microscopic ultra-early hepatocellular carcinoma lesions.

PURPOSE: This study aimed to investigate whether models built from radiomics features based on multiphase contrast-enhanced MRI can identify microscopic pre-hepatocellular carcinoma lesions. METHODS: We retrospectively studied 54 small hepatocellular carcinoma (SHCC, diameter < 2 cm) patients and 70 patients with hepatocellular cysts or haemangiomas from September 2018 to June 2021. For the former, two MRI scans were collected within 12 months of each other; the 2nd scan was used to confirm the diagnosis. The volumes of interest (VOIs), including SHCCs and normal liver tissues, were delineated on the 2nd scans, mapped to the 1st scans via image registration, and enrolled into the SHCC and internal-control cohorts, respectively, while those of normal liver tissues from patients with hepatocellular cysts or haemangioma were enrolled in the external-control cohort. We extracted 1132 radiomics features from each VOI and analysed their discriminability between the SHCC and internal-control cohorts for intra-group classification and the SHCC and external-control cohorts for inter-group classification. Five radial basis-function, kernel-based support vector machine (SVM) models (four corresponding single-phase models and one integrated from the four-phase MR images) were established. RESULTS: Among the 124 subjects, the multiphase models yielded better performance on the testing set for intra-group and inter-group classification, with areas under the receiver operating characteristic curves of 0.93 (95% CI, 0.85-1.00) and 0.97 (95% CI, 0.92-1.00), accuracies of 86.67% and 94.12%, sensitivities of 87.50% and 94.12%, and specificities of 85.71% and 94.12%, respectively. CONCLUSION: The combined multiphase MRI-based radiomics feature model revealed microscopic pre-hepatocellular carcinoma lesions.

Computed tomography-based radiomics combined with machine learning allows differentiation between primary intestinal lymphoma and Crohn's disease.

BACKGROUND: Due to similar clinical manifestations and imaging signs, differential diagnosis of primary intestinal lymphoma (PIL) and Crohn's disease (CD) is a challenge in clinical practice. AIM: To investigate the ability of radiomics combined with machine learning methods to differentiate PIL from CD. METHODS: We collected contrast-enhanced computed tomography (CECT) and clinical data from 120 patients form center 1. A total of 944 features were extracted single-phase images of CECT scans. Using the last absolute shrinkage and selection operator model, the best predictive radiographic features and clinical indications were screened. Data from 54 patients were collected at center 2 as an external validation set to verify the robustness of the model. The area under the receiver operating characteristic curve, accuracy, sensitivity and specificity were used for evaluation. RESULTS: A total of five machine learning models were built to distinguish PIL from CD. Based on the results from the test group, most models performed well with a large area under the curve (AUC) (> 0.850) and high accuracy (> 0.900). The combined clinical and radiomics model (AUC = 1.000, accuracy = 1.000) was the best model among all models. CONCLUSION: Based on machine learning, a model combining clinical data with radiologic features was constructed that can effectively differentiate PIL from CD.

Radiomics­Clinical model based on 99mTc-MDP SPECT/CT for distinguishing between bone metastasis and benign bone disease in tumor patients.

BACKGROUND: To establish a radiomics-clinical model based on 99mTc-MDP SPECT/CT for distinguishing between bone metastasis and benign bone disease in tumor patients. METHODS: We retrospectively analyzed 256 patients (122 with bone metastasis and 134 with benign bone disease) and randomized them in the ratio of 6:2:2 into training, test and validation sets. All patients underwent 99mTc-labeled methylene diphosphonate (99mTc-MDP) SPECT/CT. We manually outlined the volumes of interest (VOIs) of lesions using ITK-SNAP from SPECT and CT images. In the training set, radiomics features were extracted using PyRadiomics and selected using Least Absolute Shrinkage and Selection Operator (LASSO) regression. Then, we established three radiomics models (CT, SPECT and SPECT-CT models) using support vector machine (SVM). In addition, a radiomics-clinical model was constructed using multivariable logistic regression analysis. The four models' performance was assessed using the area under the receiver operating characteristic curve (AUC). Using DeLong test to make comparisons between the ROC (receiver operating characteristic) curves of different models. The clinical utility of the models was evaluated using decision curve analysis (DCA). RESULTS: The radiomics-clinical displayed excellent performance, and its AUC was 0.941 and 0.879 in the training and test sets. The DCA of radiomics-clinical model showed the highest clinical utility. CONCLUSIONS: The radiomics-clinical nomogram for identifying bone metastasis and benign bone disease in tumor patients was suitable to assist in clinical decision.

Whole slide imaging-based deep learning to predict the treatment response of patients with non-small cell lung cancer.

Background: This study developed and validated a deep learning (DL) model based on whole slide imaging (WSI) for predicting the treatment response to chemotherapy and radiotherapy (CRT) among patients with non-small cell lung cancer (NSCLC). Methods: We collected the WSI of 120 nonsurgical patients with NSCLC treated with CRT from three hospitals in China. Based on the processed WSI, two DL models were established: a tissue classification model which was used to select tumor-tiles, and another model which predicted the treatment response of the patients based on the tumor-tiles (predicting the treatment response of each tile). A voting method was employed, by which the label of tiles with the greatest quantity from 1 patient would be used as the label of the patient. Results: The tissue classification model had a great performance (accuracy in the training set/internal validation set =0.966/0.956). Based on 181,875 tumor-tiles selected by the tissue classification model, the model for predicting the treatment response demonstrated strong predictive ability (accuracy of patient-level prediction in the internal validation set/external validation set 1/external validation set 2 =0.786/0.742/0.737). Conclusions: A DL model was constructed based on WSI to predict the treatment response of patients with NSCLC. This model can help doctors to formulate personalized CRT plans and improve treatment outcomes.

CT-based radiomics for predicting radio-chemotherapy response and overall survival in nonsurgical esophageal carcinoma.

Background: To predict treatment response and 2 years overall survival (OS) of radio-chemotherapy in patients with esophageal cancer (EC) by radiomics based on the computed tomography (CT) images. Methods: This study retrospectively collected 171 nonsurgical EC patients treated with radio-chemotherapy from Jan 2010 to Jan 2019. 80 patients were randomly divided into training (n=64) and validation (n=16) cohorts to predict the radiochemotherapy response. The models predicting treatment response were established by Lasso and logistic regression. A total of 156 patients were allocated into the training cohort (n=110), validation cohort (n=23) and test set (n=23) to predict 2-year OS. The Lasso Cox model and Cox proportional hazards model established the models predicting 2-year OS. Results: To predict the radiochemotherapy response, WFK as a radiomics feature, and clinical stages and clinical M stages (cM) as clinical features were selected to construct the clinical-radiomics model, achieving 0.78 and 0.75 AUC (area under the curve) in the training and validation sets, respectively. Furthermore, radiomics features called WFI and WGI combined with clinical features (smoking index, pathological types, cM) were the optimal predictors to predict 2-year OS. The AUC values of the clinical-radiomics model were 0.71 and 0.70 in the training set and validation set, respectively. Conclusions: This study demonstrated that planning CT-based radiomics showed the predictability of the radiochemotherapy response and 2-year OS in nonsurgical esophageal carcinoma. The predictive results prior to treatment have the potential to assist physicians in choosing the optimal therapeutic strategy to prolong overall survival.

Dosimetric analysis and biological evaluation between proton radiotherapy and photon radiotherapy for the long target of total esophageal squamous cell carcinoma.

Objective: The purpose of this study is to compare the dosimetric and biological evaluation differences between photon and proton radiation therapy. Methods: Thirty esophageal squamous cell carcinoma (ESCC) patients were generated for volumetric modulated arc therapy (VMAT) planning and intensity-modulated proton therapy (IMPT) planning to compare with intensity-modulated radiation therapy (IMRT) planning. According to dose-volume histogram (DVH), dose-volume parameters of the plan target volume (PTV) and homogeneity index (HI), conformity index (CI), and gradient index (GI) were used to analyze the differences between the various plans. For the organs at risk (OARS), dosimetric parameters were compared. Tumor control probability (TCP) and normal tissue complication probability (NTCP) was also used to evaluate the biological effectiveness of different plannings. Results: CI, HI, and GI of IMPT planning were significantly superior in the three types of planning (p < 0.001, p < 0.001, and p < 0.001, respectively). Compared to IMRT and VMAT planning, IMPT planning improved the TCP (p<0.001, p<0.001, respectively). As for OARs, IMPT reduced the bilateral lung and heart accepted irradiation dose and volume. The dosimetric parameters, such as mean lung dose (MLD), mean heart dose (MHD), V 5, V 10, and V 20, were significantly lower than IMRT or VMAT. IMPT afforded a lower maximum dose (D max) of the spinal cord than the other two-photon plans. What's more, the radiation pneumonia of the left lung, which was caused by IMPT, was lower than IMRT and VMAT. IMPT achieved the pericarditis probability of heart is only 1.73% ± 0.24%. For spinal cord myelitis necrosis, there was no significant difference between the three different technologies. Conclusion: Proton radiotherapy is an effective technology to relieve esophageal cancer, which could improve the TCP and spare the heart, lungs, and spinal cord. Our study provides a prediction of radiotherapy outcomes and further guides the individual treatment.