Convolutional neural network in the detection of gastrointestinal tumor and tap

Abstract

At this stage, there are many studies on the pathological factors and prognosis of gastrointestinal cancer beds, but the results are different. Molecular biology and clinical studies have shown that tumor invasion and micrometastasis are likely to appear in the early stages of tumor occurrence. At present, the existing methods have poor specificity, low sensitivity, and long detection time. It is difficult to detect the occurrence or metastasis of tumors in time, and it is difficult to evaluate the efficacy in an early stage. Therefore, the detection of CTCs and their types in the peripheral blood of patients is of great significance in the early detection of tumor cell recurrence and metastasis, evaluation of prognosis and efficacy, determination of tumor molecular characteristics, and selection of appropriate individualized treatments. At the same time, the later the clinical pathological stage, the worse the prognosis, which has been widely confirmed in many studies. This study mainly explores the application of convolutional neural networks in gastrointestinal tumors and tap detection. In order to facilitate the construction, training, verification and testing of the target detection network, uses the python development language for code writing, and conducts the model training in the Ubuntu system environment. The model part is the constructed target detection network. This article uses 4 network model. Training refers to the update calculation of loss function and network. Adam optimization update network parameters, cross-entropy cost function is used to calculate category loss, and SmoothL1 is used to calculate position loss of border regression. The main function of recording is to record the loss function value and verification set accuracy data during the network training process and various detection data during the network training. During the TAP test, 20−30μl (2 drops) of blood from the end of the finger or earlobe of the subject is routinely collected and pushed into 2 large and uniform blood slices, placed in a constant temperature and humidity working environment, and placed flat. After the reaction is complete and the stain is dry, the results can be observed by microscope inspection. The colitis group was compared with the adenocarcinoma group, with statistical significance (P<0.05). This paper studies the gastrointestinal tumor detection method based on it, which will have important significance for the development and improvement of the automated auxiliary diagnosis technology of medical imaging.