Simulating federated learning for steatosis detection using ultrasound images.

We aimed to implement four data partitioning strategies evaluated with four federated learning (FL) algorithms and investigate the impact of data distribution on FL model performance in detecting steatosis using B-mode US images. A private dataset (153 patients; 1530 images) and a public dataset (55 patient; 550 images) were included in this retrospective study. The datasets contained patients with metabolic dysfunction-associated fatty liver disease (MAFLD) with biopsy-proven steatosis grades and control individuals without steatosis. We employed four data partitioning strategies to simulate FL scenarios and we assessed four FL algorithms. We investigated the impact of class imbalance and the mismatch between the global and local data distributions on the learning outcome. Classification performance was assessed with area under the receiver operating characteristic curve (AUC) on a separate test set. AUCs were 0.93 (95% CI 0.92, 0.94) for source-based partitioning scenario with FedAvg, 0.90 (95% CI 0.89, 0.91) for a centralized model, and 0.83 (95% CI 0.81, 0.85) for a model trained in a single-center scenario. When data was perfectly balanced on the global level and each site had an identical data distribution, the model yielded an AUC of 0.90 (95% CI 0.88, 0.92). When each site contained data exclusively from one single class, irrespective of the global data distribution, the AUC fell in the range of 0.34-0.70. FL applied to B-mode US images provide performance comparable to a centralized model and higher than single-center scenario. Global data imbalance and local data heterogeneity influenced the learning outcome.

Machine learning algorithm to predict response to immunotherapy in real-life settings for patients with advanced melanoma.

BACKGROUND: Melanoma is one of the most fatal forms of skin cancer. Defining relevant biomarkers to predict treatment outcome based on immune checkpoint inhibitors (ICIs) is needed in order to increase overall survival of metastatic melanoma patients (MM). OBJECTIVES: This study compared different machine learning models in terms of performance to identify biomarkers from clinical diagnosis and follow-up of MM patients, to predict treatment response to ICIs under real-life conditions. MATERIALS & METHODS: Clinical data from melanoma patients with an AJCC status of III C/D or IV, having received ICIs, were extracted from the RIC-MEL database for this pilot study. Light Gradient Boosting Machine, linear regression, Random Forest (RF), Support Vector Machine and Extreme Gradient Boosting were compared in terms of performance. The SHAP (SHapley Additive exPlanations) method was used to assess the link between the different clinical features investigated and the prediction of response to ICIs. RESULTS: RF showed the highest scores for accuracy (0.63) and sensitivity (0.64) and high scores for precision (0.61) and specificity (0.63). AJCC stage (0.076) showed the highest SHAP mean value, thus being the most suitable feature to predict response to treatment. The number of metastatic sites per year (0.049), number of months since first treatment initiation and the Breslow index (both 0.032) were less predictive, but still showed relatively high predictive power. CONCLUSION: This machine learning approach confirms that a certain number of biomarkers may enable prediction of treatment success with ICIs.