Automatic recognition of small cell carcinoma based on the self-organizing neural network

In this paper the recognition of Small Cell Carcinoma (SCC) is studied. For each type we select 128 samples for training, and randomly measure 200 cells in each sample. We introduce multi-scale morphology based on centroid coordinates to extract the boundaries of nuclei and obtain feature images of nuclei. The features of lung cancer cells are described by morphological and colorimetrical parameters, which is valuable to recognize SCC. Then the architecture of self-organizing feature mapping (SOFM) neural network is studied for recognition of SCC. The weights of the network are adjusted by self-organizing competition, and finally inputted patterns are classified. This algorithm has the advantage of parallelism and fast-convergence, and may simplify the analysis of SCC. Clinical experiment results show that the correctness ratio of this system may reach 95.3% while recognizing lung cancer cell types. Our work is significant to the pathological researches of lung cancer, assistant clinic diagnosis, and assessment of therapeutic effects. Meanwhile a software system named as SCC. LUNG is established for automatic analysis.