Lung cancer is the leading cause of cancer related deaths in the world. The survival rate can be improved if the presence of lung nodules are detected early. This has also led to more focus being given to computer aided detection (CAD) and diagnosis of lung nodules. The arbitrariness of shape, size and texture of lung nodules is a challenge to be faced when developing these detection systems. In the proposed work we use convolutional neural networks to learn the features for nodule detection, replacing the traditional method of handcrafting features like geometric shape or texture. Our network uses the DetectNet architecture based on YOLO (You Only Look Once) to detect the nodules in CT scans of lung. In this architecture, object detection is treated as a regression problem with a single convolutional network simultaneously predicting multiple bounding boxes and class probabilities for those boxes. By performing training using chest CT scans from Lung Image Database Consortium (LIDC), NVIDIA DIGITS and Caffe deep learning framework, we show that nodule detection using this single neural network can result in reasonably low false positive rates with high sensitivity and precision.
A new method for reducing the speckle in ultrasound images is introduced, which is an adaptation of Non Local Means
filter by incorporating nonlinear Gaussian for identifying the similarity of patches and restoration of pixel value. By
using this method, we are able to achieve speckle removal without using filter chains which was otherwise required for
Non linear Gaussian filters for considerable noise removal. User interaction is facilitated for controlling the amount of
noise removal and smoothing. The overall time required for computations is less and the accuracy and quality of the
images is preserved. The algorithm has been tested on phantom data as well as in vivo data. The performance measure is
evaluated based on standard evaluation parameters. On visually comparing the despeckled images, it can be found that
the structure and edge information is preserved while suppressing the speckle. Experimental results prove that this
method can be used for removing speckle in medical ultrasound images without compromising the accuracy and quality.
There are two tunable parameters in this filter. They are for controlling the amount of noise removal and smoothing. This
makes it possible for the user to adjust the amount of filtering. The filter can be easily extended to three dimensions there
by facilitating 3D volume filtering. The filter can be easily implemented in GPU (Graphics Processing Units) which
makes it possible to be used in real time particularly for volume rendering and visualization. It has been found that the
proposed Non Local Non Linear Gaussian Filtering (NL-NLG) filter exhibits the properties of edge preservation, fine
detail preservation as well as small structure preservation. At the same time it helps in the removal of speckle also. These
properties of structure enhancement, together with speckle removal increase its diagnostic capability.