摘要: Among all the respiratory cancers, the incidence of laryngeal carcinoma is second only to lung cancer, accounting for about 1% of all body tumors. At present, the most common detection method for laryngeal carcinoma is still the m... 展开 Among all the respiratory cancers, the incidence of laryngeal carcinoma is second only to lung cancer, accounting for about 1% of all body tumors. At present, the most common detection method for laryngeal carcinoma is still the method of endoscopic imaging, including white-light endoscopy and endoscopic systems with narrow band imaging (NBI). As a new optical imaging technology developed in recent years, the probe-based confocal laser endomicroscopy (pCLE) can clearly display the changes in the tissue structure at the cellular level?which can be the mainstay for judging the cell condition. It has been used in some advanced hospitals and research institutes to detect laryngeal carcinoma, and its superiority and accuracy has been confirmed. However, there is no objective evaluation system about pCLE-based image diagnosis in the current medical community. The diagnostic results and diagnostic basis solely depend on professional physicians' personal capabilities and experience. At the same time, the professional doctors studying the PCLE image are very rare, and it costs a lot to cultivate such a professional doctor. As artificial intelligence develops and integrates into various fields rapidly, the research of computer-aided diagnosis based on "AI+medical imaging" is in full swing. Hence, there is an incredible need for developing computeraided diagnosis of pCLE laryngeal carcinoma imaging. Accordingly, this paper proposes and develops an intelligent diagnostic system based on the Transformer network, which is independent of surgeons and imaging experts. This system can perform computer-aided diagnosis of laryngeal carcinoma based on pCLE images automatically and quickly with high accuracy, and determine the different phases of the pathology with accurate diagnostic basis, thereby achieving the effect of computer graphics auxiliary diagnosis. The experimental result shows that the intelligent diagnostic system proposed in this paper performs even better than pCLE professional imaging physicians with more than 5-10 years of experience in the diagnosis of different pathology of laryngeal carcinoma. And this system is also superior to the traditional Deep Convolutional neural network in various performance indicators, and its diagnostic basis has also been recognized by professional image physicians. Based on this system, an objective standard system of laryngeal carcinoma pCLE imaging diagnosis can be established, which will effectively reduce the risk of subjective diagnosis of doctors, improve diagnostic efficiency, and relieve the burden to cultivate pCLE image physicians to a certain extent. 收起