Segmentation and Recognition of Lung Adenocarcinoma Cells Based on U-Net Model

Lung adenocarcinoma poses a great threat to human health, and early diagnosis is very important for treatment. Currently, pathologists analyze and diagnose pathological cells by observing their distribution (normal cells, hyperplasia and cancer cells). So, accurate segmentation of lung cells is very important to help pathologists make diagnosis. However, it is a heavy workload to obtain information from whole slide images by human eye observation. And with the development of deep learning, its application in medical image is increasing. We can segment and recognize cells based on this technology. U-Net model, one of the most classic segmentation models, has obtained enormous number of achievements in pathological image processing. Therefore, in this paper, we propose a segmentation model for lung adenocarcinoma cells based on U-Net model. This model is trained by synthesizing pseudo-color images generated from three bands of hyperspectral images as input. We have conducted experiments on a home-made lung adenocarcinoma dataset and the results show that this method can get precise segmentation results.