Towards precision medicine based on a continuous deep learning optimized ensemble approach: A simulated prospective study of ultrasound breast mass

We developed a continuous learning system (CLS) based on deep learning and optimal integration and conducted a simulated prospective study using ultrasound images of breast masses for precise diagnoses. We extracted 629 data points and 2,235 images from 561 cases in the institution to train the model in six stages to diagnose benign and malignant tumors, pathological types, and pathological diseases. 180 cases were randomly selected from 3,098 cases from two foreign institutions. The CLS was tested with seven independent data lists in three external datasets and compared with 21 physicians, and the system’s diagnostic ability exceeded 20 physicians by training stage six. The optimal integrated method we developed is expected to achieve accurate diagnosis of breast lumps, and this method can also be used for other AI diagnosis. Overall, our findings could further promote the use of AI diagnosis in precision medicine. The dataset includes the diagnostic performance of 5 algorithms in image cropping and non-cropping, the 5 algorithms are: DenseNetImageNet121, inception_resnet_v2, nception_v3, resnet50, Xception. There are two types of images: cropped and non-cropped., the training images are stored in the train and test directories., there are two subdirectories in the train directory, namely Malignant and Benign, there are 800 images respectively. There are 2 subdirectories in the test directory, Malignant and Benign, with 165 images respectively, and 2 directories for external test data, External_test_image and External_test_cropimage, with 400 images respectively.