Image cytometry in automated cervical screening

Severe restrictions with regard to false negative rates have played a major role in the development of the LEYden Television Analysis System (LEYTAS). The present paper describes a test with a continuous series of 1500 cervical samples illustrating the accuracy of LEYTAS in a fully automated screening procedure using cell selection transformations and artefact rejection procedures. Specimen classification with a cut-off at greater than 0.3% alarms (= percentage of automatically selected objects per epithelial cells) and greater than 10 alarms, results in a false negative rate (FNR) of 0.3% (1 case out of 321 cases with severe dysplasia or more serious lesions), a false positive rate (FPR) of 13% (663 negative cases) and a rejection rate of 2.7%. Besides a machine classification, LEYTAS offers a second, machine-interaction classification of those preparations which have been declared positive by the machine. Machine-interaction involves visual evaluation of the stored images of the detected objects (alarms) and reduces the FPR from 13 to 8%. Statistical tests further demonstrate the significance of the screening results. Presently the main drawback for routine use of automated screening with LEYTAS seems to be the time consuming preparation procedure, since instrumentation has now been updated to a new, fast and user-friendly version of LEYTAS.     

Image analysis combined with quantitative cytochemistry. Results and instrumental developments for cancer diagnosis

This paper describes the application of image analysis combined with a quantitative staining method for the analysis of cervical specimens. The image analysis is carried out with the Leyden Television Analysis System, LEYTAS, of which two versions are described. LEYTAS-1 as well as LEYTAS-2 have both been designed with a high degree of flexibility and interaction facilities. A much wider range of image analysis programs is however, possible with LEYTAS-2, enabling many applications. LEYTAS-1, the earlier version, consists of a Leitz microscope with automated functions, a TV camera, the Texture Analysis System (TAS, Leitz), a four-bit grey value memory and a minicomputer (PDP 11/23). Using this instrumentation 1,500 cervical smears prepared from cell suspensions and stained with acriflavin-Feulgen-Sits have been analysed in a completely automated procedure. Image transformations working in parallel on entire fields, have been used for cell selection and artefact rejection. Resulting alarms, consisting of selected single cells and non-rejected artefacts are stored in the grey value memory, which is displayed on a TV monitor. This option allows visual interaction after the machine diagnosis has been made. The machine diagnosis was correct in 320 out 321 specimens with a severe dysplasia or more serious lesion. The false positive rate in 561 morphologically negative specimens (normal and inflammation) was 16% (machine diagnosis). Visual interaction by subtracting the visually recognized false alarms from the total number of alarms reduces the false positive rate to 11%.(ABSTRACT TRUNCATED AT 250 WORDS)     

False alarms in a slit-scan flow system: causes and occurrence rates. Implications and potential solutions.

A slit-scan technique was developed as a basis for an automated prescreening system for gynecologic cytology. A flow system based on this technique was fabricated and tested and results indicated that false alarms (misclassification of objects or events from normal specimens as abnormal) are the greatest remaining obstacle to development of an automated prescreening instrument. A dual view correlation system was fabricated to provide exact image-contour correlation in flow and permit precise determination of causes and occurrence rates of false alarms. This paper presents data from correlation analyses of 23 normal cytologic specimens. Major causes of false alarms and their implications to automated prescreening are discussed. A technique that would eliminate the majority of false alarms in flow is presented.     

Image analysis combined with quantitative cytochemistry

Summary This paper describes the application of image analysis combined with a quantitative staining method for the analysis of cervical specimens. The image analysis is carried out with the Leyden Television Analysis System, LEYTAS, of which two versions are described. LEYTAS-1 as well as LEYTAS-2 have both been designed with a high degree of flexibility and interaction facilities. A much wider range of image analysis programs is however, possible with LEYTAS-2, enabling many applications. LEYTAS-1, the earlier version, consists of a Leitz microscope with automated functions, a TV camera, the Texture Analysis System (TAS, Leitz), a four-bit grey value memory and a minicomputer (PDP 11/23). Using this instrumentation 1,500 cervical smears prepared from cell suspensions and stained with acriflavin-Feulgen-Sits have been analysed in a completely automated procedure. Image transformations working in parallel on entire fields, have been used for cell selection and artefact rejection. Resulting alarms, consisting of selected single cells and non-rejected artefacts are stored in the grey value memory, which is displayed on a TV monitor. This option allows visual interaction after the machine diagnosis has been made. The machine diagnosis was correct in 320 out 321 specimens with a severe dysplasia or more serious lesion. The false positive rate in 561 morphologically negative specimens (normal and inflammation) was 16% (machine diagnosis). Visual interaction by subtracting the visually recognized false alarms from the total number of alarms reduces the false positive rate to 11%. In LEYTAS-2, which is based on LEYTAS-1 studies, the microscope is equipped with a new type of objective, enabling the analysis of microscope fields, which are four times as large as in LEYTAS-1. The image analysis part consists of the Modular Image Analysis Computer (MIAC, Leitz) and for alarm storage an eight-bit grey value processor is used. Comparison with LEYTAS-1 shows that cell selection capacities are similar and that the speed is four times higher.In honour of Prof. P. van Duijn     

Multidimensional slit-scan detection of bladder cancer. Preliminary clinical results

A multidimensional slit-scan flow system was developed for the automated recognition of abnormal cells derived from cancer of the uterine cervix and its precursors. It provides great sensitivity in both its ability to recognize cellular abnormality and to deal with the myriad potential causes of false alarms in an automated flow system. While its initial application was the automated recognition of the spectrum of neoplasia in gynecologic cytology samples, a preliminary study was carried out using specimens obtained from the urinary bladder. Cellular material was collected by bladder irrigation and stained with the fluorochrome acridine orange. One hundred fifty-three bladder irrigation specimens, including 115 abnormal specimens containing cells derived from neoplastic lesions of the bladder epithelium, were analyzed. For the purposes of this study, abnormal specimens from the urinary bladder included specimens containing cells derived from the following lesions of the urothelium: dysplasia (atypical hyperplasia), carcinoma-in-situ, and transitional cell carcinoma, grades 1-3. Approximately 50,000 cells were analyzed for most specimens. Of the 38 presumed normal specimens (specimens containing only normal urothelial components), four were instrument classified abnormal. For the 69 specimens containing cells derived from transitional cell carcinoma, grade 1, 1-2, 2, 66 were correctly classified as abnormal while three were classified as normal.(ABSTRACT TRUNCATED AT 250 WORDS)     

Quantitative cytology of the positive region in flow sorted vaginal smears.

Fifty-one gynecologic specimens were collected from three women's hospitals and mailed in a prefixed status to our laboratory. The specimens were classified into a negative, a suspicious, a postradiation, and a positive group. After single cell dispersion the samples were stained for DNA and protein, analyzed, and sorted in the dual laser equipped Heidelberg flow analyzer sorter (HEIFAS). Particles with elevated DNA values (beyond 3.5 ploidy) and with intermediate protein values were sorted as the positive fraction directly on microscopic slides. After restaining according to Papanicolaou, they were re-evaluated cytologically and identified as tumor cells, dysplastic cells and false alarms. The latter consist of doublets and aggregates of more than two cells, binucleated cells, sperm aggregates and epithelial cells contaminated with bacteria. The different groups showed significant differences regarding the total rate of aggregates to single cells. In general, false alarms were very frequent in the positive region and impeded the statistical classification of the sample. The reduction of false alarms is a prerequisite for prescreening with flow instrumentation.     

Automated classification of bird and amphibian calls using machine learning: A comparison of methods

We compared the ability of three machine learning algorithms (linear discriminant analysis, decision tree, and support vector machines) to automate the classification of calls of nine frogs and three bird species. In addition, we tested two ways of characterizing each call to train/test the system. Calls were characterized with four standard call variables (minimum and maximum frequencies, call duration and maximum power) or eleven variables that included three standard call variables (minimum and maximum frequencies, call duration) and a coarse representation of call structure (frequency of maximum power in eight segments of the call). A total of 10,061 isolated calls were used to train/test the system. The average true positive rates for the three methods were: 94.95% for support vector machine (0.94% average false positive rate), 89.20% for decision tree (1.25% average false positive rate) and 71.45% for linear discriminant analysis (1.98% average false positive rate). There was no statistical difference in classification accuracy based on 4 or 11 call variables, but this efficient data reduction technique in conjunction with the high classification accuracy of the SVM is a promising combination for automated species identification by sound. By combining automated digital recording systems with our automated classification technique, we can greatly increase the temporal and spatial coverage of biodiversity data collection.     

Benign cellular changes in Pap smears. Causes and significance

OBJECTIVE: We reviewed consecutive cases classified as benign cellular changes (BCC) over a four-month period. STUDY DESIGN: Cases classified as BCC were retrieved from the cytology files. A search was carried out to identify the previous Pap smears and concomitant cervical biopsies. RESULTS: One thousand one hundred three cases (23% of our gynecologic smears) were classified as BCC. Ninety-two patients (8.3%) underwent concurrent cervical biopsies. Specific infections accounted for 8% of BCC cases; reactive changes accounted for 92%. Of the biopsy specimens, 8.3% had no significant pathologic change. The most common biopsy diagnoses were cervicitis (31.5%), immature squamous metaplasia (16.3%) and reserve cell hyperplasia (10.8%). Miscellaneous benign diagnoses accounted for 21.7%. Cervical intraepithelial neoplasia (CIN) 1/human papillomavirus (HPV) was present in 14% of cases. All patients with biopsy diagnoses of CIN 1 had at least two previous abnormal Pap smears. Previous biopsy reports were available for review in 127 (12%) of the 1,103 patients. Of these 127 cases, 53.5% had a previous diagnosis of CIN/HPV; 9.4% had invasive carcinoma. A benign diagnosis was reported in 36.5%. CONCLUSION: The majority of BCC cases are due to reactive and inflammatory processes. In patients with a previous history of CIN, BCC may be of some significance. In patients with no significant prior cervical abnormalities, a Pap smear classified as BCC represents a reactive process.     

A statistical analysis of prescreening alarms in a population of normal and abnormal gynecologic specimens

A multidimensional slit-scan flow system has been developed to serve as an automated prescreening instrument for gynecological cytology. Specimens are classified abnormal based on the number of cells having elevated nuclear fluorescence (alarms). An alarm region in a bivariate histogram of nuclear fluorescence versus nuclear-to-cell-diameter ratio is defined. Alarm region probability arrays are calculated to estimate the probability that an alarm falling in a particular bin of the alarm region is either from a normal or an abnormal specimen. From these arrays, a weighted alarm index is generated. In addition, summary indices are derived that measure how the distribution of alarms in each specimen compares with the average distributions for the normal and abnormal specimen populations. These indices together with current features are evaluated with respect to their utility in specimen classification using a nonparametric classification technique known as recursive partitioning. Resulting classification trees are presented that suggest information in the distribution of alarms in the bivariate histogram. In addition, they validate the features and rules currently used for specimen classification. Recursive partitioning appears to be useful for multivariate classification and is seen as a promising technique for other applications.     

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.     

Analysis of machine-selected cells with an image analysis system in normal and abnormal cervical specimens

An analysis has been performed of visual diagnostic criteria used in cervical cytology applied to machine selected cells in relation to automated classification based on variables, which can be recorded in an image system with automated cell search and segmentation, feature extraction and classification. A 98% accuracy could be obtained with the choice of the most ideal statistical methods for discrimination and the use of the most powerful variables recorded in the image system when compared with consensus of the visual diagnoses based on established cytological criteria for diagnosis of cancer and precancer of the cervix uteri. The most powerful discriminatory variables in the image system (of 17 recorded) for discrimination between normal and abnormal epithelial cells were, in addition to nuclear extinction, cytoplasmic extinction and cytoplasmic shape. It is concluded that the visual classification of cervical cells is highly accurate with experienced observers and that imaging microscopes can be trained to nearly equal this accuracy with appropriate statistical methods of discrimination. The problem of creating fully automated systems, however, also requires the inclusion of even more effective discriminatory variables and also the solution of such problems as automatic cell search, segmentation, artifact rejection, feature extraction, classification and electronic stability in order to become cost-effective.     

Imaging systems for correlation of false alarms in flow.

False alarms, arising from a variety of sources, are the greatest remaining obstacle to development of an automated prescreening system for gynecologic cytology. This paper describes two correlation systems under development at the University of Rochester and discusses their utilization in the study of false alarms in slit-scan cytofluorometry. Both systems permit imaging of objects in flow and correlation between images and corresponding slit-scan contours. Correlation systems will permit a detailed study of false alarm causes and aid in the search for new features to assist in their recognition.     

Cytophotometric analysis of corresponding cytological and histological cervical intraepithelial neoplasia grade III specimens

Cytophotometric analysis of cervical intraepithelial neoplasia grade III (CIN III) was performed in 22 cytological smears (CS) and in 22 corresponding cytospin specimens retrieved from selected areas of paraffin-embedded tissues (PEC). The average time interval between cytological and histological diagnosis was 6 weeks. CIN III nuclei in CS and PEC specimen were Thionin-Feulgen stained and digitized. Beside the visual classification of DNA ploidy patterns, the 2.5c and 5c exceeding rates and the specimen mean and standard deviation values of 21 photometric features were also analyzed. It was shown that, although there was a significant correlation between DNA ploidy patterns in corresponding PEC and CS specimen, the DNA patterns were dissimilar in eight of 22 cases. The DNA index, as represented by 2.5c and 5c exceeding rates, was significantly higher in the CS specimen. High-resolution cytophotometric analysis of cell nuclei in CS and PEC specimens showed significant differences for a large number of nuclear photometric features. These findings can possibly be explained by differences in selection of CIN III cells from CS and PEC specimens and by differences between fixation procedures as used for the two techniques. It was concluded that cytophotometric data of CS and PEC specimens representing CIN III lesions should not be regarded as interchangeable.     

Accuracy of liquid-based Pap tests: comparison of concurrent liquid-based tests and cervical biopsies on 782 women with previously abnormal Pap smears

OBJECTIVE: To evaluate the diagnostic performance of a liquid-based Pap test, the ThinPrep Pap test (TP) (Cytyc Corp., Boxborough, Massachusetts, U.S.A.), by comparing concurrent TP and cervical biopsy results on 782 patients who were referred for colposcopy because of previously abnormal conventional Pap smears (CPs). STUDY DESIGN: The ability of TP diagnoses of atypical cells of undetermined significance (ASC-US) and squamous intraepithelial lesions (SILs) to predict biopsy diagnoses of cervical intraepithelial neoplasia (CIN) was analyzed using chi2 and McNemar tests. RESULTS: The rate of agreement between diagnoses of SIL by TP and CIN by biopsy was 74.7%. ASC-US accounted for 16.0% of TP diagnoses. ASC-US had biopsy diagnoses of CIN 1 in 60% and CIN 2/3 in 12.8% of cases. For TP diagnosis of low grade SIL, biopsy diagnoses of CIN 2/3 were found in 13.5% of cases. For TP diagnoses of ASC-US and higher, the proportions of TP and cervical biopsies in comparable diagnostic categories were statistically significant (p < 0.001), with TP having sensitivity of 89.4% and positive predictive value of 89.7% for the detection of CIN. The false positive rate for TP was 8.1%, but rescreening confirmed the presence of abnormal cells in 51 of 63 (81.0%) cases of ASC-US or higher having negative biopsies. TP had a false negative rate of 8.3% and negative predictive value of 61.3%. Rescreening showed that most (77.6%) of the false negative TP specimens failed to have abnormal cells on the slides. CONCLUSION: For patients having previously detected cervical abnormalities by CP, concurrent TP demonstrated the following: (1) that it has high diagnostic accuracy for SIL, (2) that ASC-US was diagnostically equivalent to LSIL, and (3) that false negative TP for SIL can be attributed primarily to sampling rather than cytotechnologists' screening errors.     

Machine vision automated species identification scaled towards production levels

Computer‐automated identification of insect species has long been sought to support activities such as environmental monitoring, forensics, pest diagnostics, border security and vector epidemiology, to name just a few. In order to succeed, an automated identification programme capable of addressing the needs of the end user should be able to classify hundreds of taxa, if not thousands, and is expected to distinguish closely related and hence morphologically similar species. However, it remains unknown how automated identification methods might handle an increase in data quantity, be it in reference imagery or taxonomic diversity. We sought to test the scalability of an automated identification method in terms of the number of reference specimens used to train the classifier and the number of taxa into which the classifier should assign unknown specimens. Is there an optimal number of reference images, where the cost of acquiring more images becomes greater than the marginal increase in identification success? Does increasing taxonomic diversity affect identification success, whether negatively or positively? In order to test the scalability of the automated insect identification enterprise, we used a sparse processing technique and support vector machine to test the largest dataset to date: 72 species of fruit flies (Diptera: Tephritidae) and 76 species of mosquitoes (Diptera: Culicidae). We found that: (i) machine vision methods are capable of correctly classifying large numbers of closely related species; (ii) when the misclassification of a specimen occurs at the species level, it is often classified in the correct genus; (iii) classification success increases asymptotically as new training images are added to the dataset; (iv) broad taxon sampling outside a focal group can increase classification success within it.     

A selective mapping algorithm for computer analysis of voided urine cell images

One of the fundamental targets of the automated image analysis of cytologic preparations is the reduction of computer classification errors due to cells or other objects that do not lend themselves to image segmentation or that have morphologic features that may mislead the cell classification schemes. In prior work from this laboratory, the achievement of this goal was attempted by hierarchical analysis of sequential microscopic objects at high resolution. This paper reports on the successful development and implementation of an automated "selective mapping algorithm" that selects cells at low power for further analysis and eliminates a large proportion of unwanted "objects." The algorithm classifies the objects and extracts appropriate features from a 256 X 240 digital image obtained via a 10 X planachromatic objective. The five-node binary tree classifier used in this triage is described. The algorithm was trained and tested initially on 501 visually classified microscopic "objects," resulting in a correct acceptance rate of 61.3% and correct rejection rate of 81.3%. The selective mapping algorithm was subsequently integrated into the video-based image analysis system constructed at the Montefiore Medical Center for the diagnostic evaluation of sediments of voided urine. The algorithm was then tested on ten cytocentrifuge preparations for a preliminary evaluation of its performance. Up to 100 "objects" per case were selected by the algorithm for further classification by the computer at high power. Of the 810 "objects" selected by the selective mapping algorithm, 344 (42.5%) were classified by the computer at high resolution as cells of diagnostic value ("WELL" cells) and 466 were rejected.(ABSTRACT TRUNCATED AT 250 WORDS)     

Comparative study of cytology and punch biopsy in cervical intraepithelial neoplasia during pregnancy. A preliminary report

To elucidate the accuracy of cytology in diagnosing cervical intraepithelial neoplasia (CIN) during pregnancy, cytologic screenings for uterine cervical cancer in pregnant women were reviewed for a five-year period. Of the 967 pregnant women screened, abnormal cytologic findings were recorded for 15 (1.6%). Only nine of these were subsequently examined by colposcopy and punch biopsy, which demonstrated CIN in all cases, for an incidence of documented CIN during pregnancy of 0.93%. Including two referral cases also examined by colposcopy and biopsy, cytology and histology agreed on the degree of CIN in four cases and disagreed by one degree in four cases, by two degrees in two cases and by three degrees (mild dysplasia versus carcinoma in situ) in one case. Review of the specimens from these cases did not readily explain the poor concordance between cytology and punch biopsy; some findings suggest that overestimation of the punch biopsy sample may be the explanation.     

Capturing the songs of mice with an improved detection and classification method for ultrasonic vocalizations (BootSnap)

House mice communicate through ultrasonic vocalizations (USVs), which are above the range of human hearing (>20 kHz), and several automated methods have been developed for USV detection and classification. Here we evaluate their advantages and disadvantages in a full, systematic comparison, while also presenting a new approach. This study aims to 1) determine the most efficient USV detection tool among the existing methods, and 2) develop a classification model that is more generalizable than existing methods. In both cases, we aim to minimize the user intervention required for processing new data. We compared the performance of four detection methods in an out-of-the-box approach, pretrained DeepSqueak detector, MUPET, USVSEG, and the Automatic Mouse Ultrasound Detector (A-MUD). We also compared these methods to human visual or ‘manual’ classification (ground truth) after assessing its reliability. A-MUD and USVSEG outperformed the other methods in terms of true positive rates using default and adjusted settings, respectively, and A-MUD outperformed USVSEG when false detection rates were also considered. For automating the classification of USVs, we developed BootSnap for supervised classification, which combines bootstrapping on Gammatone Spectrograms and Convolutional Neural Networks algorithms with Snapshot ensemble learning. It successfully classified calls into 12 types, including a new class of false positives that is useful for detection refinement. BootSnap outperformed the pretrained and retrained state-of-the-art tool, and thus it is more generalizable. BootSnap is freely available for scientific use.     

Attention parameters in visual search tasks in various human age groups

The visual search performance was investigated in two children (5 and 7-years-old) and three adult groups (15, 20--and 35 and up to 60-years-old). The results of adults were similar among themselves and therefore were taken as a control set for children group results. We used 5 sorts of visual tasks, one difficult task and four easy tasks for adults with a target differing from counter-irritation. Children's results were worse than adults' those: the search time was longer and children made plenty of errors (false alarms end misses of targets). We suggest that children's attention system was less effective than adults' one in result of frontal, parietal and temporal cortical structure activation deficit. It follows that attention system is completely developed by 15 years of age in humans.     

Bladder cancer diagnosis by computer image analysis of cells in the sediment of voided urine using a video scanning system

A video-based computerized semiautomated image analysis system was applied to the diagnostic evaluation of 119 sediments of voided urine: 103 from patients with a broad variety of neoplastic and nonneoplastic disorders of the lower urinary tract and 16 normal controls. Each specimen was presented to the machine as a single cytocentrifuge preparation, preserved in 2% Carbowax in 50% ethanol and stained-by the Papanicolaou method. Five hundred sequential "objects" were scanned within an area of 9 sq mm on each slide. "Objects" of no diagnostic value, such as dirt, debris, inflammatory cells, cell clusters, poorly preserved cells, etc., were eliminated from the final diagnostic analysis by a computer-based hierarchic triage system. The final specimen classifier was based on the cell images identified by the computer as well-preserved normal (NEG), atypical (ATY I), suspicious (ATY II) and malignant (POS) cells. For specimen classification by computer, the four categories of "abnormal," "inadequate," "acellular" and "negative" were defined. For high-grade tumors, the performance of the specimen classifier was generally comparable to the visual diagnosis. The specimen classifier unexpectedly identified twice as many low-grade papillary urothelial tumors as abnormal than did the visual analysis. Several false "alarms" were recorded by computer in patients with benign prostatic hypertrophy and prostatic carcinoma, some of whom had atypical urothelium. One of the 16 negative controls was misdiagnosed by the computer as abnormal. The possibility that the video system recognizes nuclear abnormalities not perceived by the human eye is being investigated further. The details of the computer analysis are reported, and the value of the system is discussed. The system appears to be promising as a future laboratory instrument, although it requires further extensive testing.