The shortwave infrared (SWIR) shows a significantly higher level of contrast between different types of tissues compared to the Vis-NIR. This contrast can be further enhanced by using hyperspectral imaging (HSI). Unlike conventional imaging methods, each HSI pixel contains a full high-resolution spectral signal. Complex computational approaches are required to analyze HSI-measured optical alterations. Many of these approaches including machine learning and deep learning algorithms have been developed in the field of geospatial imaging. We adapted and optimized algorithms used in remote sensing to build an image processing platform IDCube that processes biomedical data and identifies tissues abnormalities in SWIR.
Machine learning methods using convolution neural networks (CNN) are one technique which has played an important role in many biomedical applications. Here we apply the techniques to the detection of retinal blood vessels in ophthalmological images. Since the eye is one of the most important sensory organs in the human body, it is critical to diagnose diseases in the early stages. Early symptoms of various diseases like glaucoma, diabetic retinopathy, and cardiovascular diseases can be detected via the structure of the blood vessels of the retina. Studying the retinal blood vessel structures and network requires blood vessel segmentation. Deep learning has been used for the last five years and achieved state-of-the-art performance. More specifically, the latest development has been implemented to segment blood vessels more efficiently by using the u-net architecture. In the process, we used forward convolutional long short-term memory (convLSTM) to combine the feature map of the encoding path and the corresponding decoding path in lieu of simple concatenation in the skip connection of the u-net. We also used a connected convolution layer to the last layers of the encoding path to obtain more diverse features. The images used were from the DRIVE database and preprocessing was performed to obtain more accurate results. We achieved 95% accuracy, a precision of 91%, a sensitivity of 52%, and a specificity of 99%. Even with a low sensitivity of 52%, all the major blood vessels are found successfully.
Compression algorithms have been implemented for k-means and k-means++ clustering and applied to thermographic images. The overall algorithm has four stages and are the same for the two algorithms except for the initialization of the centroids. The compression ratio and quality are primarily dependent on the number of clusters used for the algorithm. A MATLAB GUI was developed to run the algorithms and a comparison has been performed with subjective evaluations and objective RMS error, peak SNR and compression ratio metrics. The average compression ratio was 1.3 and 1.6 for the k-means and k-means++ clustering respectively. The k-means++ clustering provides subjectively better visual results than the standard k-means clustering.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.