No abstract

No abstract

Pneumonia has become one of the main causes of human death. However, it is a tall order to efficiently and accurately diagnose pneumonia for clinicians.Therefore, A novel method based on anchor-free detection framework is proposed to automatically locate lung opacities on chest radiographs in this study.We conducted extensive sets of experiments on the dataset of the Radiological Society of North America (RSNA) pneumonia detection challenge from the Kaggle competition. The results show superior performances for our method compared with previous studies. The best method achieved 52.9% in average precision (AP) and 97.5% in average recall (AR). For better interpretability of the results, visualization techniques are applied to provide visual explanations for our method. The visualization of these randomly selected samples shows that the method has excellent performance for lung opacity detection. Our method achieves better discriminative results and is suitable for the pneumonia diagnosis.

Ophthalmic diseases afflict many people, and can even lead to irreversible blindness. Therefore, the search for effective early diagnosis methods has attracted the attention of many researchers and clinicians. At present, although there are some ways for the early screening of ophthalmic diseases, the early screening of fundus images based on deep learning is generally favored by the medical community due to its non-contact characteristic, non-invasive characteristic and high recognition accuracy. However, the generalization performance of a common model and cross-domain identification is usually weak due to different collection equipment, race, and patient conditions. Although the existing fundus image recognition technology has achieved some results, the effect is still in the cross-domain problem and is not satisfactory. In this paper, a cross-domain fundus image recognition framework based on deep neural networks with data enhancement is proposed. First, the ResNeXt101 model is chosen as the basic framework. Second, some data enhancement methods and focal loss are used to improve recognition performance. Finally, the results of experiment show that the final score of the framework is improved by about 10% using ordinary data enhancement methods and focal loss. In summary, the method proposed in this paper can effectively solve the problem of poor generalization ability for cross-domain early fundus screening and can provide inspiration and ideas for future related works.

No abstract