UNSUPERVISED SEGMENTATION OF GALL BLADDER LESIONS USING HIDDEN MARKOV RANDOM FIELD MODEL Koushik Chakraborty 1 *, Arunava De 2 , Sudhir K. Sharma 3 1 Jayoti Vidyapeeth Women’s University, Jaipur,India 2 K L University, Guntur,India 3 Jaipur National University, Jaipur,India koushik215@gmail.com Abstract: Gallbladder cancer is an uncommon cancer prevalent in some geographical locations such as South America, central and eastern Europe, Japan and northern India. If it is detected early then the patient can be cured by removing the gallbladder, portions of liver and lymph nodes. There are no specific symptoms of this disease and the cancer remains undetected till it has spread to adjoining organs. Hence the early detection of gallbladder lesions may save crucial lives. MR image of gallstones may result in early detection of gallbladder lesions. This article proposes to detect gallbladder lesions using artificial intelligence and soft computing techniques. Hidden Markov Random Field Model is used for segmentation of Gall-bladder lesions.Expert medical opinion is required to conclude whether the lesions have developed cancer. Keywords: Gall Bladder Lesions, Hidden Markov Random Field Model 1. Introduction To better understand disease and to quantify its evolution we use Magnetic Resonance imaging. Manual identification of lesion border is a time taking process. It is also prone to observer variability. We require fully automated and reproducible method to correctly segment the lesions and also those should be free of observer variability. Markov models have shown effective results for a variety of phenomena. The use of these models has increased in the fields of finance, economics, ecology, communications, signal and image processing. Hidden Markov model is effective in treating the problem of segmentation. The hidden data which models the desired segmented image may follow an example of a field, tree or a chain. HMM’s are also used to treat inverse problems in imaging such as noise removal etc. We define the segmentation of lesions in Gall bladder as a pixel labelling problem. Hidden Markov Random Field (HMRF) is used to segment the MR image into foreground and background labels. Ising model is used a prior to ensure that the foreground and background components are coherent. 2. MRI Data Acquisition of Gall Bladder We have taken STIR FRFSE Resp Trig Fat SAT MR sequences (STIR-Short-T1 Inversion Recovery, FRFSE- Fast Recovery Fast Spin Echo, Respiratory Triggered Fat Saturated). STIR stands for Short-T1 Inversion Recovery and is used to nullify the signals from FAT. There is uniform fat suppression by STIR and independent of magnetic field in-homogeneities. It is better than other fat saturation methods such as “spectral-fat-sat” for abdomen and pelvic areas. Respiratory Triggering is a type of imaging involving respiratory motion. During expiration MR images are acquired. The scan time is dependent on patient’s breathing patterns. Fat Sat saturates fat protons prior to image acquisitions. We have acquired 22 MR images in this mode of image acquisition using slice thickness of 0.5 mm and resolution of 512 × 512. 3. Related Works The mathematical theory of Markov Process was named after Andrei Markov in the early twentieth century [1]. The theory of Hiddden Markov Models (HMM’s) was developed by Baum in 1960’s [2]. Entropy Maximization using Particle Swarm Algorithm was used by [3] for segmenting the MR image of brain. Ref [4] used Entropy Maximization using Grammatical Swarm algorithm for segmenting lesions of human brain.Hybrid Particle Swarm optimization with Wavelet mutation based segmentation and progressive International Journal of Pure and Applied Mathematics Volume 117 No. 19 2017, 343-348 ISSN: 1311-8080 (printed version); ISSN: 1314-3395 (on-line version) url: http://www.ijpam.eu Special Issue ijpam.eu 343