Adaptable ring, Bayes theorem and dual gradient analysis: An approach to automatic shape detection, skin lesion segmentation and focus detection respectively for medical image analysis

“Vision-based applications are increasing with the increase in technology and are making a huge impact in this global civilization. A vision-based measurement approach for pill shape detection was developed for medical pill identification, a focus detection method was developed for digital dermoscopy, and a machine learning approach was used for automatic lesion segmentation, which assists in automatic melanoma diagnosis. Rapid and accurate pill identification is needed by medical and law enforcement personnel during emergencies. But, one of the main parameters in pill identification is the shape of the pill. An Adaptable-Ring approach describes a novel technique to accurately detect the complex pharmaceutical pill shapes using measurements derived from a superimposed adaptable ring. It is shown that these measurements suffice to successfully classify the shapes of the pills currently in the Pillbox database (U.S. National Library of Medicine, 2014). Though developed for the domain of pharmaceutical pill shapes, this method can be applied to other industrial applications. Identification of an unfocused image is essential during automatic diagnosis of melanomas because, when an unfocused image of a lesion is used in a real-time diagnostic application, it might lead to a false diagnosis. So, a dual gradient analysis method was developed to successfully identify out-of-focus dermoscopy images. Also, robust lesion segmentation is required for better automatic melanoma identification. A Bayesian model was used to find the probabilities of skin in a given image after obtaining data from just 90 training images. Finally, the non-skin area is considered as the lesion, and post processing is done accordingly to obtain the lesion from the image. Finally, this method was tested on 1636 dermoscopy images to obtain the accuracy rates”--Abstract, page iv