![Input: Vector of objects representing s dimensions, in
our case it will be an image pixel and each pixel having
three dimensions, RGB.
k: number of clusters
Output: set of k clusters that minimizes the sum of
distance error
Algorithm steps-
1. Initialize each pixel with random positive weights
{Wqk} between [0,1]
2. Assign standard initial weights for each qth feature
vector for all clusters via
Cancer Detection using FCM
Algorithm
11](https://image.slidesharecdn.com/seminar3476-170520191304/85/Lung-Cancer-Detection-using-Fuzzy-Clustering-and-Artificial-Neyron-Network-12-320.jpg)
Lung Cancer Detection using Fuzzy Clustering and Artificial Neyron Network
- 1. Lung Cancer Detection using Artificial Neural Network and Fuzzy Clustering Guided by – Dr. Sandhya Arora Presented by – Riya Tirole
- 2. Overview Basics • Artificial Intelligence • McCulloch-Pitts Model of Neuron • Artificial Neural Networks • Hopfield Neural Networks • Fuzzy Clustering Cancer Detection using • Hopfield Neuron Network (HNN) • Fuzzy C-Means Clustering (FCM) 1
- 4. ● Also called linear threshold gate ● Neuron with n inputs and 1 output McCULLOCH-PITTs NEURON MODEL 3
- 5. Artificial Neural Networks (ANN) ● Parameters - ● Interconnection pattern between different layers of neuron ● Weights of these interconnections (updated during learning process) ● Activation function that converts neuron's weighted input to output activation. 4
- 6. ● Form of recurrent ANN, and comes under unsupervised learning ● Neurons numbered as 0, 1, ... i, j, …n and weight of edge between them wij Hopfield Neural Networks (HNN) Input Output 5
- 7. Also called soft clustering ALGORITHM: ● Choose number of clusters ● Assign randomly to each point coefficients for being in a cluster ● Repeat until algorithm has converged • Compute centroid, ck for each cluster • For each point, compute its coefficient of being in cluster Fuzzy Clustering 6
- 8. ● Abnormal cell growth ● 39.6% of all men and women are detected with cancer ● Out of these 19% is only lung cancer ● Cancer detected in advanced stages when survival chance is low Cancer 7
- 9. ● Input : RGB Components of image ● Sensitive to intensity variation and hence can detect cytoplasm overlapping ● Neural network has N x M neural units, where, row – pixel, column – cluster ● Image classified into N pixels, P features among M classes ● Energy function (E) is given as, Where Rkl is Eucledian distance and Vkl is output of kth neuron Cancer Detection using HNN 8
- 10. 1.Initialize input of neurons to random values 2.Obtain new values and assign them to one of the class 3.After formation of classes, calculate centroid. 4.Solve differential equation to update neuron values 5.Repeat step 2 to 4 until convergence. Algorithm for HNN 9
- 11. ● Out of 1000 sputum images the algorithm could segment components with 97% of accuracy. ● Took less time to achieve desired result ● Experimentally, took less than 120 iterations to reach desired segmentation result in 36 seconds. Conclusion for HNN 10
- 12. Input: Vector of objects representing s dimensions, in our case it will be an image pixel and each pixel having three dimensions, RGB. k: number of clusters Output: set of k clusters that minimizes the sum of distance error Algorithm steps- 1. Initialize each pixel with random positive weights {Wqk} between [0,1] 2. Assign standard initial weights for each qth feature vector for all clusters via Cancer Detection using FCM Algorithm 11
- 13. 3. Standardize the weights over k = 1,…,K for each q to obtain Wqk, via 4. Compute new centroids C(k), k = 1,….,K 5. Update the weights {Wqk } via 6. If there is change in the input, repeat from step 3, else terminate. 7. Assign each pixel to a cluster based on the maximum weight.12
- 14. ● When given 1000 sputum images it also segments it into different components but with less precision. ● Less sensitive to intensity variation, therefore all the cytoplasmic region in one cluster whatever k is given. ● Detect only part of nuclei ● Experimentally, took less than 50 iterations to reach desired segmentation result in 10 seconds. Conclusion for FCM 13
- 15. Comparison (a) Original raw image stained with blue dyes, (b) and (c) the segmentation results for the image in (a) by using HNN and FCM, respectively. (d) The filtered image. (e) And (f) show the segmentation results for the filtered image in (d) by using HNN and FCM, and by fixing the cluster numbers to three, respectively. (g) And (h) the results by fixing the cluster numbers four, respectively.14
- 16. Comparison The learning error wave forms of HNN and FCM during the segmentation process, for the blue cells image in the image in slide above 15
- 17. ● It was noticed that HNN was more accurate and reliable than FCM in all the cases like extracting nuclei and cytoplasmic region. ● Moreover, FCM is not sensitive to intensity variation as segmentation error at convergence is larger in FCM than HNN. ● Therefore, HNN will be used in Computer Aided Diagnosis (CAD) systems for early detection of lung cancer. Conclusion 16
- 18. ● Fatma Taher, Rachid Sammounda, “Lung Cancer detection by using Artificial Neural Networks and Fuzzy Clustering Methods” February 19-22, 2011 IEEE GCC Conference and Exhibition (GCC), Dubai, United Arab Emirates ● R. Sammouda, N. Niki, H. Nishitani, S. Nakamura, and S. Mori, “Segmentation of Sputum Color Image for Lung Cancer Diagnosis based on Neural Network,” IEICE Transactions on Information and Systems. vol. 8, pp. 862-870, August, 1998 ● http://wwwold.ece.utep.edu/research/webfuzzy/docs/k k-thesis/kk-thesis-html/node12.html ● https://www.cdc.gov/cancer/international/statistics.htm References 17
- 19. THANK YOUTHANK YOU for being such an AMAZING AUDIENCE...AMAZING AUDIENCE... 18
- 20. 19