Convolutional neural network automatic segmentation method and system for mammary molybdenum target data set

Abstract

The invention discloses a convolutional neural network automatic segmentation method and system for a mammary molybdenum target data set, which can obviously reduce model parameters and improve the practicability while ensuring the precision of a deep learning model on a mammary molybdenum target small data set. The method comprises the following steps: pre-training a convolutional neural big network on a mammary molybdenum target big data set; performing model compression on the trained convolutional neural large network by adopting attention transfer and knowledge distillation methods to obtain a convolutional neural small network; and carrying out fine tuning on the convolutional neural small network on the mammary molybdenum target small data set.

Description

technical field [0001] The disclosure relates to an automatic segmentation method and system of a deep fully convolutional neural network based on an attention transfer-based knowledge distillation method applied to a small breast mammography data set. Background technique [0002] Breast cancer is the cancer with the highest incidence rate in women. Studies have shown that breast cancer accounts for 29% of cancer incidence and 15% of cancer mortality in women. Early diagnosis of breast cancer is crucial to the survival of patients. Among various breast cancer early screening techniques, mammography has the advantages of low dose, high sensitivity, simplicity and convenience. Radiologists often misdiagnose cancer or overdiagnose cancer due to differences between entities and observers during mammography analysis. Therefore, computer-aided diagnosis is very important as a pre-screening method for disease diagnosis. significance. At the same time, due to the large difference...

AI Technical Summary

Problems solved by technology

However, a robust deep learning model requires a large amount of data from different data sources as a training set, and the amount of high-quality medical data that is often available is small. Therefore, how to train a deep learning model on small medical data is more robust. It is also one of the main factors that hinder the implementation of this method

In addition, in order to further squeeze the learning ability of the deep learning model, the design of the model structure tends to have deeper layers and more parameters. Practicality forms a contradiction

Images