A Feed Forward Neural Network For Recognition Of Lung Nodules In Ct Images

Abstract

The best precise imaging method for determining the presence and stage of lung cancer is considered to be computed tomography (CT). On a chest X-ray, nodules are a legitimately frequent abnormality: one out of every 500 chest X-rays reveals newly diagnosed nodules. From helical CT scans, we suggested a computer-aided diagnostic (CAD) method to detect small-size lung knots, which range in size from 1 mm to 6 mm. An insignificant, curving (parenchymal knot) or caterpillar-shaped (juxtapleural nodule) wound in the lungs is known as a pulmonic knot. Because they each have a higher radio-density than the lung parenchyma, they appear snowy in images. Lung knots may indicate a lung cancer, and identifying them early on improves patient survival rates. The best precise imaging technique for finding knots is thought to be CT. However, because there is so much data in each study, analysis becomes difficult. This suggests that a human radiologist could have missed a knot. The proposed CAD method aims to reduce omissions and shorten the time needed for radiologist review of the picture. Our method classifies nodule items from non-nodule objects using a three-layer Feed Forward Neural Network with directed knowledge based on back-propagation technique and CLAHE, an alternative to Histogram Equalization that reduces the noise amplification. The technique was tested on Windows after being built in Matlab. Simple graphic user interface is supplied for convenient control.