24/7 writing help on your phone
Lung cancer is a leading disease which results in huge death rates which has compared using combined death rates among prostate, colon and breast cancer. Lung cancer detection in early stage is essential to improve the survival rate of lung cancer patients. The overall five year survival rate of lung cancer for an advanced stages is only 16% but the early stage increases to 70% However, in an early stage, lung cancer manifests itself in the form of pulmonary nodules which refer as lung tissue abnormalities.
These are roughly spherical with round opacity on a CT scan. Due to huge number of image slices at a CT scan, larger workload arises among radiologists for a diagnostic reading in a short time and it prone to errors. To overcome such challenge for an radiologist, we have developed various computer aided detection (CAD) systems in last recent years.
In an existing CAD system, it consists of two stages: candidate nodule detection and false positives reduction.
During first stage, upon an assumed expected morphology of the nodules, large number of candidate regions are generated using intensity based imaging features. In a second stage, false positives results from the first stage are reduced using a classifier. This stage is very important for a successful use of CAD in clinical practices. If there remains too many false positives, the performance of detecting true positive regions may be degraded and it increase in unnecessary follow-up examinations. To reduce false positives as much as possible, hand-crafted features such as Scale Invariant Feature Transform(SIFT), Local Binary Patterns (LBP) and Histogram of Oriented Gradients (HOG) are extracted from the candidate regions and used to train a classifier for nodule classification.
However, under such a condition, the performance of a CAD system relies heavily on the intermediate results of the
image processing tasks for reliable features whereas the selection, integration and optimization of hand-crafted features are not an easy task, it is also an important issue in image classification and also has an important effect on the performance of a CAD system.
Recently,remarkable progresses have been obtained in the field of computer vision due to the recent revival of deep convolutional neural networks (CNN) and the availability of large-scale annotated datasets(i.e ImageNet). Compared to the conventional machine learning methods, deep learning requires no hand-tuned feature extractor and learns imaging representations directly from large volumes of data.
👋 Hi! I’m your smart assistant Amy!
Don’t know where to start? Type your requirements and I’ll connect you to an academic expert within 3 minutes.get help with your assignment