Scientific progress is based on the establishment of scientific integrity, and academic misconduct has been a stumbling block in the way of scientific career1. Among them, image forgery occupies a certain position, and with the development of technology, such as the emergence of powerful functions of Photoshop, ImageJ and other graphics software, while providing convenience for researchers, it is also abused by some authors who try to manipulate images, making the means of image forgery more technical and covert2. In the face of such a critical situation, it is urgent and important to prevent image fraud, which requires enriching techniques for deep image verification. The easiest way to solve this problem is to make the work of researchers accessible to non-specialists. For example, when evaluating liver disease, researchers should provide slides of the entire liver, rather than partial zooms of selected areas, and use technology to show all lesion slides of the entire liver, so that reviewers, editors, and even laypeople can understand the presence of lesions and the effectiveness of treatment. Here, artificial intelligence (AI) technology is one such technological tool.
Artificial intelligence (AI) is a technology that enables computer to mimic human intelligence. The explosion of machine learning, especially deep learning, has led to a shift in the way we live our daily lives3. Among other things, in the field of medicine, AI has enormous potential, whether in supporting clinical diagnosis, patient management or treatment planning4. Given the sheer volume that characterizes medical data itself, it provides room for continuous advancement in deep learning. It is currently excelling in driving various imaging-related fields, including but not limited to radiology5, oncology6 and digital pathology7.
Nowadays, AI has been able to automatically detect tumor tissues, thus reducing the workload of pathologists8. In addition, AI may be able to capture more of the histological features in predicting tumor staging than pathologists, and this is particularly noticeable in gastrointestinal and liver tumors9. Studies have shown that tumor-infiltrating lymphocytes are closely related to the prognosis of gastric cancer patients, and tumor-infiltrating lymphocytes on pathological tissues can be detected by Convolutional Neural Network (CNN) models, thus providing a good predictive prospect for tumor prognosis10. In hepatocellular carcinoma (HCC), AI models can predict survival after tumor resection more accurately than a multifaceted composite total score11.In particular, AI has been shown to infer molecular and genetic changes in cancer tissues from histological digital sections9. This indicates that the application of artificial intelligence in pathology provides a good environment for the development of precision treatment of digestive system tumors.
Liver lesions reflect changes in the function and metabolism of the organism caused by external stimuli or its own diseases12. By observing morphological and pathological changes in the liver, we can grasp the onset and progression of various diseases and the body's response to drug efficacy and toxicity. Liver lesions in animal models mainly include inflammatory diseases13, metabolic diseases14, and proliferative diseases15. In previous studies, we established that thioacetamide(TAA)-induced abnormal proliferation of intrahepatic bile duct epithelium leads to bile duct intraepithelial neoplasia, which is a precancerous from of bile duct cancer16. We have used this model to intervene with aspirin, herbs, and hydrogen-rich water with positive results. In the experiment with hydrogen-rich water intervention17,our algorithmic model clearly shows the location and amount of liver fibrosis, visualizes the differences between the two treatment options, and provides objective quantitative indicators. However, there are two other main features of this animal model that this algorithm does not address: one is the surface state of the liver and the other is the difference between fibrosis and neoplasia.
For these animal liver models, we have developed a new deep learning network for liver disease segmentation to identify different pathological features of liver lesions, with the aid of artificial intelligence to visualize the lesion sites and quantify the results more intuitively.