FIGURE 2 from Artificial Intelligence–Assisted Cancer Status Detection in Radiology Reports

Training and inference of BERT model. A, Illustration of data preprocessing, data partition with balanced stratification, hyper parameter optimization (HPO) using cross validation (CV) technique and fine-tune training using CV method. B, Training loss curve while fine-tuning BERT model on multiclass classification of cancer status task along with exact accuracy measured after every epoch on evaluation data split. C, An example of inferencing of model trained on multilabel classification of imaging scan site. A radiology report excerpt (highlighted in green) is passed as input to the model, the model's output with selected label choices and corresponding label probabilities are shown. The visual from SHAP explainer model for label “Chest” is provided below model output.