Neural network application in the discrimination of benign from malignant gastric cells

OBJECTIVE: To investigate the potential value of morphometry and neural networks for the discrimination of benign from malignant gastric lesions. STUDY DESIGN: One thousand cells from 19 cases of cancer, 19 cases of gastritis and 56 cases of ulcer were selected as a training set, and an additional 4,000 cells from the same cases of cancer, gastritis and ulcer were used as a test set. Images of routinely processed gastric smears stained by the Papanicolaou technique were analyzed by a custom-made image analysis system. RESULTS: Application of the neural network gave correct classification in 96% of benign cells and 89% of malignant cells. CONCLUSION: The results indicate that the use of neural networks and image morphometry may offer useful information concerning the potential of malignancy in gastric cells.     

A comparative study of three variations of the learning vector quantizer in the discrimination of benign from malignant gastric cells

A prospective study was undertaken to investigate the potential value of morphometry and artificial neural networks (ANN) for the discrimination of benign and malignant gastric lesions. Two thousand five hundred cells from 23 cases of cancer, 19 cases of gastritis and 58 cases of ulcer were selected as a training set, and an additional 8524 cells from an equal number of cases of cancer, gastritis and ulcer were used as a test set. Images of routine processed gastric smears stained by the Papanicolaou technique were processed by a custom image analysis system. The application of the learning vector quantization (LVQ) classifier enabled correct classification of > 97% of benign cells and > 95% of malignant cells, obtaining an overall accuracy of > 97%. This study presents the capabilities of ANN, and also indicates that ANN and image morphometry may offer useful information on the potential of malignancy in gastric cells.     

Textural analysis in the diagnosis of benign and malignant breast cells

OBJECTIVE: To study the discriminatory capacity of textural variables to classify the nuclei of breast tumor cells as benign or malignant, using a statistical approach. STUDY DESIGN: Image analysis techniques were used to automatically segment nuclei of cells obtained by fine needle aspiration and Papanicolaou stained. The sample comprised 95 cases of malignant lesions and 47 cases of benign lesions (approximately 25 nuclei per case), and 27 textural variables were measured. Two methods were used to analyze the data: classification and regression trees (CART) and discriminant analysis. RESULTS: The variance in gray levels was the most decisive variable in the CART analysis, correctly classifying 57% and 97% of benign and malignant cases, respectively. Discriminant analysis yielded the best results, correctly classifying 79% and 85% of benign and malignant cases, respectively. CONCLUSION: The classifier obtained by a statistical approach to the textural analysis of Papanicolaou-stained nuclei did not prove useful for diagnostic discrimination. Staining techniques that are not chromatin specific are highly variable, and other features have proven more effective with this type of staining.     

Discrimination of benign common nevi from malignant melanoma lesions by use of features based on spectral properties of the wavelet transform

OBJECTIVE: To evaluate the possibilities of describing and discriminating common nevi and malignant melanoma tissue with features based on spectral properties of the Daubechies 4 wavelet transform. STUDY DESIGN: Images of common nevi and malignant melanoma were dissected in square elements. The wavelet coefficients were calculated inside the square elements. The diagonal coefficients and related power spectra were used for further analysis. The analysis results served as guide for the selection of features, including standard deviations of wavelet coefficients inside the frequency bands and the energy of the frequency bands. These features describe properties of the frequency bands, representing information on different scales. To test the usefulness of the features for discrimination, a study set of 80 cases was classified by classification and regression trees analysis. The set was divided into a training set and a test set. RESULTS: In the case of benign common nevi, the energies of the lower frequency bands and higher, whereas malignant melanoma tissue shows more variability of the coefficients in higher-frequency bands. The influence on the detail properties of the images was studied by suppression of coefficients with low values, which are concentrated mainly in higher-frequency bands. In the case of benign common nevi the main information is contained in 15% of the coefficients and in the case of malignant melanoma, in 39%. The results of classification show a clear-cut difference between the cases. The classification correctly classified 95.78% of nevi elements and 94.22% of melanoma elements in the training set and 100% of cases of benign nevi and 80% of cases of malignant melanoma in the test set. CONCLUSION: Features based on the wavelet power spectrum contain sufficient information for differentiation between common nevi and malignant melanomas.     

Diagnostic assessment of cutaneous melanoma and common nevi using tissue counter analysis

OBJECTIVE: To test the applicability of tissue counter analysis to the diagnostic discrimination of cutaneous malignant melanoma and benign common melanocytic nevi. STUDY DESIGN: Forty cases each of melanoma and nevi were consecutively sampled. After creation of a learning set based on 20 cases each, discriminant analysis of background versus tissue elements, tumor versus other tissue elements and benign versus malignant tumor elements was performed. The discriminant functions were used to assess the amount of benign and malignant tumor elements in each case. RESULTS: In the learning set, discriminant analysis facilitated recognition of 99.6% of tissue versus background, 90.7% of tumor versus other tissue components and 85.6% of malignant versus benign tumor elements. In the whole set, the percentage of malignant tumor elements was 8.7 +/- 7.7 (range, 0.1-29.6) for benign nevi and 77.4 +/- 16.2 (43.6-99.4) for malignant melanoma. Based on these measurements, the correct diagnosis could be established in all cases (chi 2 > .0001). CONCLUSION: Tissue counter analysis may be a useful method for diagnostic purposes in histopathology.     

Infrared spectroscopy of exfoliated cervical cell specimens. Proceed with caution

OBJECTIVE: To determine whether diagnostic information may be recovered from the infrared spectra of exfoliated cell specimens by using a novel spectral feature extraction method, in conjunction with linear and quadratic discriminant analysis, for spectral classification. STUDY DESIGN: Over 800 infrared spectra were included in the study, with corresponding clinical diagnoses based upon cytology and, when available, histology reports. Three sets of classification trials were carried out with the aim of distinguishing the spectra corresponding to normal specimens from CIN 1, 2 and 3. For each of these three cases, the procedure was to: (1) develop a set of provisional classification models using only a "training" subset of the spectra, and (2) test each provisional model by its ability to correctly predict the diagnoses on the basis of the remaining spectra. RESULTS: For optimal classification trials, training set classification accuracies were 68% for normal/CIN 1, 73% for normal/CIN 2 and 81% for normal/CIN 3; for the corresponding test sets the classification accuracies were 60%, 60% and 67%, respectively. CONCLUSION: The infrared spectra of exfoliated cervical cells carry information regarding the presence or absence of dysplasia, and that information is recoverable--albeit imperfectly at this stage--from the spectra of "real life" cell preparations.     

Evaluation of texture features in spatial and frequency domain for automatic discrimination of histologic tissue

OBJECTIVE: To investigate the applicability of different texture features in automatic discrimination of microscopic views from benign common nevi and malignant melanoma lesions. STUDY DESIGN: In tissue counter analysis (TCA) the images are dissected into square elements used for feature calculation. The first class of features is based on the histogram, the co-occurrence matrix and the texture moments. The second class is derived from spectral properties of the wavelet Daubechie 4 and the Fourier transform. Square elements from images of a training set are classified by Classification and Regression Trees analysis. RESULTS: Features from the histogram and the co-occurrence matrix enable correct classification of 94.7% of nevi elements and 92.6% of melanoma elements in the training set. Classification results are applied to individual test set cases. Discriminant analysis based on the percentage of "malignant elements" showed correct classification of all nevi cases and 95% of melanoma cases. Features derived from the wavelet and Fourier spectrum showed correct results for 88.8% and 79.3% of nevi and 85.6% and 81.5% of melanoma elements, respectively. CONCLUSION: TCA is a potential diagnostic tool in automatic analysis of melanocytic skin tumors. Histogram and co-occurrence matrix features are superior to the wavelet and the Fourier features.     

Canonical analysis of cells in normal and abnormal cervical smears

Over 4,000 cells from 105 normal and 96 abnormal uterine cervical scrapes were prepared according to the UCLA monolayer procedure, stained by a routine Papanicolaou method and visually classified by two cytopathologists and a technologist into seven classes: parabasal, metaplastic, mild dysplasia, moderate dysplasia, severe dysplasia, carcinoma in situ and invasive carcinoma. Canonical analysis was used to correlate effects-coded class membership variables with 23 cell features derived from digital image analysis. In general, nuclear texture measures derived from linear combinations of run-length correlations along with features derived from a Markov transitional probability matrix provided the best predictors of cell class. After cells were divided into benign (moderate dysplasia or less) and malignant (severe dysplasia or worse) groups, discriminant analysis correctly classified 84% of the benign cells and 91% of the malignant cells.     

Potential of radial basis function neural networks in discriminating benign from malignant lesions of the lower urinary tract

OBJECTIVE: To investigate the potential value of morphometry and neural network tools for discriminating benign from malignant nuclei and lesions of the lower urinary tract. STUDY DESIGN: The study group consisted of 33 cases of lithiasis, 41 cases of inflammation, 66 cases of benign hyperplasia of the prostate, 4 cases of carcinoma in situ, 48 cases of grade 1 transitional cell carcinoma of the bladder (TCCB) and 123 cases of grade 2 and 3 TCCB. Images of routinely processed voided urine smears stained by the Giemsa technique were analyzed by a custom image analysis system. Analysis of the images gave a data set of features from 31,158 nuclei. A radial basis function (RBF)-type neural network was employed to discriminate benign from malignant nuclei, based on the extracted morphometric and textural features. Subsequently a second RBF classifier was employed to discriminate benign from malignant cases. The nuclei from 156 randomly selected cases (50% of total cases) was used as a training set, and the nuclei from the remaining 159 cases made up the test set. Similarly, in an attempt to discriminate at the patient level, the same 156 cases were used to train an RBF classifier; the remaining 159 cases were used for the test set. The cases used for training and testing the 2 classifiers (nuclear and patient level) were the same for the 2 kinds of classifiers. RESULTS: Application of the RBF classifier permitted the correct classification of 93.64% of benign nuclei and 85.61% of malignant, giving an overall accuracy of 84.45%. At the patient level the RBF classifier permitted an overall accuracy of 94.97%. These results were on the test sets. CONCLUSION: The role of nuclear morphologic features in the cytologic diagnosis of lower urinary tract alterations was confirmed by the results of this study. The observed overlap in feature space indicates that the nuclear characteristics do not form strictly separate clusters; that fact explains the difficulty morphologists have with reproducible identification of nuclei from the lower urinary tract. Application of RBF offers good classification at the nuclear and patient level and promises to become a powerful tool for everyday practice in the cytologic laboratory.     

Computerized interactive morphometry as an aid in the diagnosis of pleural effusions

A morphometric study of cytologic preparations from patients with benign and malignant (mesothelioma and carcinoma) pleural effusions is reported. The routine cytologic smears from these specimens were studied with a new system of video-based computerized interactive morphometry (CIM) that allows the measurements of real-time images of cell profiles by the simple procedure of touching the two extreme points of a diameter of interest on a touch-sensitive screen. For each cell, the nuclear profile diameter (NPD) and the cytoplasmic profile diameter (CPD) are measured and categorized into classes with 2-microns intervals; the NPD/CPD ratio is also calculated. The mean NPD is calculated for the specimen after measurement of 100 cells. The data were interpreted by two independent methods: a statistical method of discriminant analysis that classifies the lesions as benign, carcinoma or mesothelioma and provides a probability statement of membership in a particular diagnostic class and an ad-hoc algorithm that categorizes the effusions as benign or malignant based on hierarchic analysis. A data base derived from study of the first 24 cases was constructed and utilized for the test classification of the second 24 cases, which were treated as specimens of unknown diagnosis. The discriminant analysis correctly classified 21 of the 24 test cases into their proper diagnostic groups. The algorithm for a computer-generated pathologic diagnosis correctly identified 47 of the 48 cases as benign or malignant. The technical advantages of video-based CIM over the existing morphometric methods are discussed.     

Contextual analysis complements single-cell analysis in the diagnosis of breast cancer in fine needle aspirates

Image analysis techniques were used to characterize individual nuclei of cells and entire clusters of cells in hematoxylin-and-eosin-stained smears of fine needle aspirates of the breast to determine the ability of these techniques to distinguish benign from malignant cases. Analysis of the individual nuclear features showed significant differences in nuclear area, shape (bending energy), texture and integrated darkness between benign and malignant samples. Analysis of the clusters demonstrated that the benign clusters were fewer in number, more cellular (average gray level) and larger than malignant clusters. A statistical classifier was constructed to test the discriminatory accuracy for benign and malignant cases. Good discrimination was found for both the individual nuclei and the clusters when analyzed separately, although a few cases were misclassified by each type of analysis. When combined, the two classifiers achieved a completely accurate classification. This suggests the complementary nature of high-resolution single-cell analysis and the more global cluster analysis techniques.     

Computerized interactive morphometry of brushing cytology specimens

Forty-two bronchial brushing cytology specimens were evaluated by a video-based computerized interactive morphometry (CIM) system with an interactive peripheral consisting of a touch-sensitive screen mounted over a high-resolution video monitor. The system was programmed to allow a trained observer to rapidly measure nuclear and cytoplasmic profile diameters of randomly selected cells and to calculate their nuclear-cytoplasmic ratios. The specimens included 13 cytologic slides with no malignant cells present, 14 with non-small cell carcinoma cells and 15 with small cell carcinoma cells. The cases were divided into two groups: a training set composed of slides with "known diagnosis" and a test set of slides with "unknown diagnosis". A data set was constructed with the measurements from the cases with "known diagnosis," and an algorithm that allowed the classification of cases by hierarchical analysis was developed. The data was also analyzed with statistical methods of classificatory discriminant analysis. Utilizing the information in the data base, the slides with "unknown diagnosis" were classified individually; all cases were correctly classified by the procedures. Potential applications of CIM in cytology are discussed.     

Evaluation of contextual analysis for computer classification of cervical smears

A procedure for automated analysis of cervical smears has been implemented in an image cytometry system. Smears are described exclusively in terms of global and contextual information extracted by pattern-recognition algorithms and represented by a vector of proportions of cellular object types. Linear discriminant functions, based on a Fisher criterion, are derived to classify smears with a cross-section of diagnoses into two broad categories, normal and abnormal. Results obtained from 83 smears indicate 78% correct classification. In contrast to most automated systems, good classification results were obtained in normal smears with benign changes caused by inflammation and with postmenopausal atrophia and in abnormals with mild dysplasia. These findings suggest that contextual analysis may be sensitive to subtle changes in cellular morphology and to progressive patterns of dysplasia. When used with standard isolated cell analysis, contextual analysis may provide additional complementary information for automated cervical prescreening.     

Prediction of various grades of cervical neoplasia on plastic-embedded cytobrush samples. Discriminant analysis with qualitative and quantitative predictors.

The purpose of this study was to investigate whether discrimination into five groups of various grades of cervical preneoplasia and neoplasia is possible using discriminant analysis models. Data were analyzed for 242 cases diagnosed as either slight dysplasia (n = 50), moderate dysplasia (n = 50), severe dysplasia (n = 50), carcinoma in situ (n = 50) or invasive carcinoma (n = 42) and consisted of qualitative and quantitative features of cells derived from a repeat sample taken from the ectocervix as well as the endocervix using Cytobrushes. The samples were embedded in plastic, and thin sections were prepared, resulting in a monolayer of cut nuclei. The percentages of expected correct prediction were obtained by using 10,000 double cross-validation samples; the mean percentage of correct prediction into five groups using cross-validation was 65% (in the original analysis, 72%) and into two groups (dysplasia versus carcinoma in situ and invasive carcinoma) was 91% (93%). The results reflect group discrimination potential; we do not claim reliability of prediction for an individual patient. The patients were not a representative sample of the population; to investigate whether groups of patients could be discriminated on the basis of both qualitative and quantitative features, the data analyzed contain an almost equal number of observations in each of the five groups. The results indicate that features do not classify the cases in the same way; the discriminant analyses suggest that quantitative features play an important role in the discrimination of dysplasia from carcinoma cases, while the majority of the qualitative features are important in discrimination within the three dysplasia groups.     

Computerized microscopic analysis of prostatic fine needle aspirates. Comparison with breast aspirates

Computerized image analysis was employed to analyze fine needle aspiration smears of the prostate and breast using both high-resolution images of individual cells and medium-resolution images of scenes and clusters (contextual analysis). A linear discriminant analysis was used to demonstrate the computer's ability to discriminate between benign and malignant categories for both types of tissue. Correct classification as benign or malignant using contextual analysis was achieved in 22 of 26 prostatic aspirates and in 15 of 18 breast aspirates, as determined by comparison with histology. The addition of high-resolution single-cell analysis resulted in correct classification of 24 of 26 prostatic aspirates and all breast aspirates. For virtually all features, the distinction between benign and malignant was more subtle for prostatic than for breast tissue. The data indicate that contextual analysis may be less effective as an adjunct to high-resolution single-cell microscopy of prostatic specimens than it is for breast specimens.     

Discriminant analysis of image karyometric variables in benign and malignant melanocytic skin lesions

OBJECTIVE: To perform a quantitative analysis to identify which of 7 nuclear morphometry-related variables are of diagnostic value in distinguishing benign from malignant melanocytic skin lesions. STUDY DESIGN: At the Institute of Pathology, University of Nis, formalin-fixed, paraffin-embedded skin biopsies from 23 cases of benign nevi (18 intradermal and 5 junctional) and 25 cases of primary nodular malignant melanomas were retrieved. Specimens were routinely stained with hematoxylin and eosin and analyzed using a computer-assisted interactive image analysis system. Nuclear area, equivalent diameter, volume of equivalent sphere, perimeter, mean chord, circularity and integrated optical density were estimated after manual editing of binary images. RESULTS: In univariate analysis, 6 features were found to be significantly different between the benign and malignant groups (P < .0001); all measured nuclear variables (except circularity) were higher in malignant melanomas. No significant differences were found among lesions with respect to nuclear shape. Using discriminant function analysis, a correct diagnosis was achieved in 95.8% of benign nevi cases and 84.0% of malignant melanoma cases. The best discriminant variable was nuclear area. CONCLUSION: Image analysis is diagnostically relevant to the evaluation of melanocytic lesions of the skin. The area of the nucleus appeared to have potential for differentiating benign from malignant tumors and can be estimated in the course of routine histology.     

Morphometric, densitometric and flow cytometric criteria for the automated classification of thyroid lesions

An automated classification of 73 thyroid lesions using a logical and mathematical approach was attempted. Densitometric, morphometric and flow cytometric parameters were used in Fisher linear discriminant functions to separate goiters or normal thyroids from adenomas and from carcinomas; the combination of this approach with binary discrimination improved the initial classification to a final efficiency of 81%. This approach, which is useful for classifying individual cells, was thus insufficient for classifying these cases. Analysis of the individual parameters showed that thyroid lesions were mainly in the near-diploid region. Two G0G1 populations were present in both benign and malignant lesions and were particularly frequent (50%) in atypical invasive follicular adenomas, probably related to the additional presence of an invasive clone. Near-triploid peaks were associated with malignancy as well as with high proliferative indexes. Nuclear and nucleolar sizes were larger in carcinomas; however, the percentage of the nucleolar area in the nucleus was greater in adenomas and nodular adenomatous goiters. A corrected staining index correlated with the nuclear size and the ploidy of abnormal cells (r = .50), being higher in malignant lesions.     

Image analysis of low magnification images of fine needle aspirates of the breast produces useful discrimination between benign and malignant cases

Image analysis of low magnification images of fine needle aspirates of the breast produces useful discrimination between benign and malignant cases
Fine needle aspirates of the breast (FNAB) ( n =362; 204 malignant, 158 benign), prepared by cytocentrifuge methods and stained by the Papanicolaou technique, were analysed using a semi‐automated image analysis system at a low magnification which precluded resolution of nuclear detail. The measured parameters were integrated optical density, fractal textural dimension, number of cellular objects (single cells and contiguous groups of cells), distance between cellular objects (mean, s.d., skewness and kurtosis), area of cellular objects (mean, s.d., skewness, kurtosis) and the nearest neighbour statistic. The cases were divided into a 200‐case training set and a 162‐case test set. Analysis was performed by logistic regression and the multi‐layer Perceptron type of artificial neural network. Logistic regression and the neural network produced similar performances with a sensitivity of 82–83%, specificity 85% and a positive predictive value for a malignant result of 85%. A non‐parametric analysis of all the predictor variables showed that all except the mean area of cellular objects and the s.d. of this measurement were significant discriminants ( P <0.05), but most were highly interrelated and this was reflected in the selection of only three predictor variables by forward and backward conditional logistic regression. This study shows that much diagnostic information is present in low power views of FNAB, and that image analysis could form the basis of a semi‐automated decision‐support aid.     

Detection of bladder cancer using proteomic profiling of urine sediments

We used protein expression profiles to develop a classification rule for the detection and prognostic assessment of bladder cancer in voided urine samples. Using the Ciphergen PBS II ProteinChip Reader, we analyzed the protein profiles of 18 pairs of samples of bladder tumor and adjacent urothelium tissue, a training set of 85 voided urine samples (32 controls and 53 bladder cancer), and a blinded testing set of 68 voided urine samples (33 controls and 35 bladder cancer). Using t-tests, we identified 473 peaks showing significant differential expression across different categories of paired bladder tumor and adjacent urothelial samples compared to normal urothelium. Then the intensities of those 473 peaks were examined in a training set of voided urine samples. Using this approach, we identified 41 protein peaks that were differentially expressed in both sets of samples. The expression pattern of the 41 protein peaks was used to classify the voided urine samples as malignant or benign. This approach yielded a sensitivity and specificity of 59% and 90%, respectively, on the training set and 80% and 100%, respectively, on the testing set. The proteomic classification rule performed with similar accuracy in low- and high-grade bladder carcinomas. In addition, we used hierarchical clustering with all 473 protein peaks on 65 benign voided urine samples, 88 samples from patients with clinically evident bladder cancer, and 127 samples from patients with a history of bladder cancer to classify the samples into Cluster A or B. The tumors in Cluster B were characterized by clinically aggressive behavior with significantly shorter metastasis-free and disease-specific survival.     

慢性腹痛患儿C~(13)-尿素呼气试验与胃镜检查的临床意义

目的了解慢性腹痛患儿幽门螺杆菌(H.pylori)的感染状态及幽门螺杆菌感染患儿内镜下表现的特点。方法应用C13尿素呼气试验,对905例以慢性腹痛为主要症状的患儿进行检测,对C13呼气试验阳性者进行电子胃镜检查。结果905例慢性腹痛患儿中H.pylori呈阳性185例(20.44%),随年龄增长,其H.pylori阳性率升高,学年组已达高峰。对H.pylori阳性者进行胃镜检查结果显示十二指肠隆起病变47例占25.40%,结节性胃炎41例占22.1%,慢性浅表性`胃炎38例占20.5%,结节性胃炎伴十二指肠隆起病变23例占12.43%,十二指肠球部溃疡23例占12.4%。胃溃疡7例,占3.7%(其中包括1例复合性溃疡),结节性胃炎伴十二指肠炎6例,占3.2%。结论H.pylori感染为小儿慢性腹痛的主要原因之一,也是导致慢性胃炎及消化性溃疡的主要原因之一。C13尿素呼气试验方便,快速,无痛苦,无放射性,是一较好的H.pylori检测方法;对既有消化道症状同时C13呼气试验阳性者进行胃镜检查能够协助临床诊断及治疗。