Data processing: database and file management or data structures – Database design – Data structure types
Reexamination Certificate
1994-06-03
2002-10-08
Shah, Sanjiv (Department: 2172)
Data processing: database and file management or data structures
Database design
Data structure types
C707S793000
Reexamination Certificate
active
06463438
ABSTRACT:
TECHNICAL FIELD OF THE INVENTION
The present invention relates, in general, to automated image recognition, and more particularly, to a neural network-based image recognition system for cancerous tissue cell detection.
BACKGROUND OF THE INVENTION
Bladder cancer is among one of the top causes of cancer-related deaths in the United States. In 1993, approximately 52,000 cases of bladder cancer were reported in the United States and 10,000 deaths were attributed to this disease. Early detection is essential in minimizing the risks involved with bladder cancer. The detection of bladder cancer is traditionally performed through cystoscopy, through the quantitation of plasma components of urine, or through detailed analysis of stained bladder cells obtained from urine or from a bladder wash.
Cystoscopy is the visual inspection of the bladder using a fiber optic device. Normally performed by a urologist, cystoscopy is discomforting to the patient and exposes the patient to the risks and costs associated with surgery. Further, cystoscopy only detects cancerous cells after the tumor has progressed to an advanced stage.
Significant progress in the detection and isolation of bladder tumor specific antigens has linked bladder cancer with an elevation of normal protein components in the plasma or urine of cancer patients. Thus, bladder cancer may be detected by identifying the abnormal presence of materials in the bladder cells. Since these tests are non-invasive, they could be routinely utilized to test those in high risk groups for early symptoms of bladder cancer. However, an approach using serum and plasma related components in urine appears to have limited usefulness in the early detection of bladder cancer as many of these identical components are also present in increased concentrations in urine from patients with non-neoplastic disease.
Another promising approach involves the analysis of bladder cells obtained from urine or a bladder wash. In this process, bladder cells are extracted from urine or a bladder wash. They are then prepared using conventional staining techniques such as the Papanicolaou technique for highlighting the region of interests in the sample cells. Conventionally, these cells are visually inspected for signs of cancer. Typically, after a cyto-technician has screened the sample cells, the final diagnostic is performed by an expert cytopathologist. This process is labor intensive because it requires exhausting inspection of thousands of cells. Naturally, the tedium and fatigue imposed upon the technician and the cytopathologist result in a high false negative rate.
Due to the vast amount of data to be processed, automation of the bladder cancer cell detection process is quite desirable. Various techniques have been proposed for the automated detection of cancer. Predominately, these prior attempts have relied on techniques such as feature extraction, template matching and other statistical or algorithmic methods. For instance, Melder and Koss described a decision tree representing the hierarchical classification scheme to classify extracted features from the triage of objects encountered in the urinary sediment. Karl K. Melder & Leopold G. Koss, “Automated Image Analysis in the Diagnosis of Bladder Cancer,” 26 Applied Optics 16, 3367 (1987). Melder and Koss discussed the use of step-wise linear discriminant analysis in which features were automatically selected for the discriminant functions based on the pooled covariance matrix of more than sixty (60) cell features. Christen, et al., discussed the application of a linear discriminant model from the SPSS/PC+ statistical package to the classification of cancerous cells. Christen, et al., “Chromatin Texture Features in Hematoxylin and Eosin-Stained Prostate Tissue,” 16 Analytical and Quantitative Cytology and Histology, 16, 383 (1993).
Recently, artificial neural networks have been applied to the cancer detection process. This step is a logical extension of the pattern recognition capability of artificial neural networks. Kunihiko Fukushima, “Neural Network Model for Selective Attention in Visual Pattern Recognition and Associative Recall,” 26 Applied Optics 23, 4985 (1987); Dwight D. Egbert, et al., “Preprocessing of Biomedical Images for Neurocomputer Analysis,” IEEE Int'l Conference on Neural Networks I-561 (Jul. 24-27, 1988).
A variety of neural network topologies have been experimented with. By way of illustration, some of these neural network models include the Perceptron, described in U.S. Pat. No. 3,287,649 issued to F. Rosenblatt and further described in M. Minsky and S. Papert, “Perceptrons, An Introduction to Computational Geometry,” (MIT Press 1988); the Hopfield Net, described in U.S. Pat. Nos. 4,660,166 and 4,719,591 issued to J. Hopfield; “The Hamming Network and Kohonen Self-Organizing Maps,” described in R. Lippman, “An Introduction to Computing with Neural Nets,” IEEE ASSP Magazine, April 1987 at 4-22; D. E. Rumelhart, G. E. Hinton and R. J. Williams, “Learning Internal Representations by Error Propagation,” in 1 Parallel Distributed Processing 318-362 (D. E. Rumelhart, et al. eds., 1986); and G. O. Stone, “An Analysis of the Delta Rule and the Learning of Statistical Associations,” in 1 Parallel Distributed Processing 444-459 (D. E. Rumelhart, et al. eds., 1986).
A particularly robust type of neural network is referred to as the back-propagation network. The training process for back-propagation type neural networks starts by modifying the weights at the output layer. Once the weights in the output layer have been altered, they can act as targets for the outputs of the hidden layer, changing the weights in the hidden layer following the same procedure as above. This way the corrections are back-propagated to eventually reach the input layer. After reaching the input layer, a new test is entered and forward propagation takes place again. This process is repeated until either a preselected allowable error is achieved or a maximum number of training- cycles has been executed.
Due to the sheer number of computational cycles in the training process, the computation of the activation function is crucial to the performance of the neural network. A traditional back-propagation neural network utilizes the non-linear sigmoid function as the activation function in its neurons. The effectiveness of traditional back-propagation neural networks is limited by the fact that the training procedure does not guarantee convergence to the global minima.
Traditional back-propagation neural networks have been applied to bladder cells in Ciamac Moallemi, “Classifying Cells for Cancer Diagnosis Using Neural Network,” 6 IEEE Expert 6, 8 (1991). Moallemi describes the application of a conventional neural network in the classification of noisy particles versus cell images, including cancerous and non-cancerous bladder cells. However, Moallemi does not teach the detection of malignant cells using a neural network.
The application of neural networks to the classification of cytological specimens is discussed in U.S. Pat. No. 4,965,725 to Rutenberg. Rutenberg describes the use of a two-staged classifier system. The first classifier is a statistical classifier which identifies cell nuclei of interest by measurement of their integrated optical density, or nuclear stain density, defined as the sum of the pixel gray values for the object. Rutenberg discloses that, compared to normal cells, malignant cells tend to possess a larger, more densely staining nucleus. Based on the data provided by the primary classifier, Rutenberg further employs a neural network as a secondary classifier for evaluating the nucleus and its surrounding cytoplasm based on the observation that the ratio between the nucleus and the cytoplasm is an important indicator for malignant cell classification. However, Rutenberg does not utilize other predictive information such as the pgDNA value of a cell.
One limitation with conventional back-propagation network is that it imposes considerable computational demands under its iterative gradient descent me
Ashenayi Kaveh
Hu Ying
O'Dowd Gerard J.
Veltri Robert W.
Fulbright & Jaworski
Shah Sanjiv
Urocor Inc.
LandOfFree
Neural network for cell image analysis for identification of... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Neural network for cell image analysis for identification of..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Neural network for cell image analysis for identification of... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2998605