NEURAL NETWORKS FOR THE ANALYSIS OF SMALL PULMONARY NODULES

Purpose: Small pulmonary nodules can be readily detected by computed t omography (CT), The goal of this defection is to diagnose early lung c ancer as the five year survival at this early stage is over 70% in con tradistinction to the overall 5-year survival of around 10%. Critical to the efficacy of CT for early lung cancer detection is the ability t o distinguish between benign and malignant nodules. We explored the us efulness of neural networks (NNs) to help in this differentiation. Met hods: CT images of 28 pulmonary nodules, 14 benign and 14 malignant, e ach having a diameter less than 3 cm were selected, All were sufficien tly malignant in appearance to require needle biopsy and surgery. The statistical-multiple object detection and location system (S-MODALS) N N technique developed for automatic target recognition (ATR) was used to differentiate between these benign and malignant nodules. Results: S-MODALS was able to correctly identify all but three benign nodules. S-MODALS classified a nodule as malignant because if looked similar to other malignant nodules. It identified the most similar nodules to di splay them to the radiologist. The specific features of file nodule th at determined its classification were also shown, so that S-MODALS is not simply a ''black box'' technique but gives insight into the NN dia gnostics. Conclusion: This initial evaluation of S-MODALS NNs using pu lmonary nodules whose CT features were very suspicions for lung cancer demonstrated the potential to reduce the number of biopsies without m issing malignant nodules, S-MODALS performed well, but additional opti mization of the techniques specifically for CT images would further en hance its performance. (C) Elsevier Science Inc., 1997.