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)     

DNA cytophotometry of voided urine sediment. Comparison with results of cytologic diagnosis and image analysis

Measurements of nuclear DNA were performed in urothelial cells in 54 Feulgen-restained cytocentrifuge preparations of voided urine previously studied visually and with an image analysis system. The study included 30 patients with bladder tumors of various grades, 9 patients with prostatic disease and 15 control samples from normal donors. A number of additional control measurements were performed, including measurements in tissue samples of the 30 bladder tumors corresponding to the cytologic samples. It was documented that DNA can be measured in most urinary sediments. The diagnostic performance of the image analysis system reflected the DNA patterns in 47 of the 54 cases. In several instances, particularly in cases of prostatic disease, the image analysis system recognized abnormal DNA patterns in the absence of significant morphologic abnormalities in the urothelial cells. In seven cases, the image analysis findings failed to conform with the DNA patterns. The reasons for these surprising results are discussed, and future modifications of the image analysis system are proposed.     

The effect of filtration on the loss of abnormal cervical cells in specimen preparation for automated cytology

The UCLA monolayer procedure is used to produce a preparation of cervical epithelial cells for automated cytology. We investigated the nylon mesh filtration step of the procedure to determine if it altered the proportion of abnormal epithelial cells in the specimen. The proportion of such cells was found to be significantly lower (P less than 0.05) in the final specimen retained by a 10 micron nylon mesh filter than in the filtrate, which has been routinely discarded in our studies. The filtrate specimen is contaminated by large numbers of polymorphonuclear leukocytes, however, making it unusable for automated cytology at present. The implications of our findings for the cost effectiveness of classifier performance are considered, and the implications of the results for further improvements in specimen preparation are discussed.     

Assessment of fine needle aspiration as a screening test for occult prostatic carcinoma

As part of an ongoing study of objective parameters of prognostic value in prostatic carcinoma, a routine procedure was developed to aspirate all prostates prior to surgery. These targets were different from those of other workers in the field of prostatic fine needle aspiration (FNA), who generally advocate that FNA be confined to suspicious nodules. The aspirations were performed by a large group of practicing urologists who had no special training in prostatic FNA except for guidelines provided by their peers and information available in the literature. This approach permitted an assessment of the performance of FNA as a screening test rather than as a diagnostic procedure. During the period from January 1983 to February 1987, 1,683 patients had prostatic FNAs performed (plus subsequent histologic study). The following diagnoses were rendered: "inadequate/scanty specimen" in 625 cases (37%), "negative/atypical" in 844 cases (50%) and "suspicious/positive" in 214 cases (13%). Histologic examination showed stage A1 prostatic adenocarcinoma in 18 patients. The cytologic diagnoses on these 18 patients were inadequate/scanty in 3 (17%), negative/atypical in 13 (72%) and suspicious/positive in 2 (11%). Of the 214 patients with a positive/suspicious diagnosis by FNA, the diagnosis of prostatic carcinoma was confirmed by tissue evidence in 200; the other 14 patients had either no evidence of prostatic carcinoma on surgical biopsy (needle biopsy/transurethral resection/suprapubic prostatectomy) or had no surgical biopsy. Eight of the 14 patients developed clinical evidence of carcinoma, 1 died of urinary bladder carcinoma and 1 was lost to follow-up. In the remaining four patients, there is still no evidence of prostatic carcinoma after about one-and-one-half years of follow-up. These results indicate that (1) specialized training is required in order to obtain adequate smears by prostatic FNA; (2) prostatic FNA is not a good screening technique for detecting stage A1 prostatic carcinoma; and (3) a positive diagnosis by prostatic FNA, even when not confirmed by tissue biopsy, is still an indication of disease.     

Application of alpha-methylacyl coenzyme A racemase immunohistochemistry in the diagnosis of prostate cancer: a review

Alpha-methylacyl coenzyme A racemase (AMACR) is a recently discovered enzyme protein that has been shown to be increased at both the mRNA and protein levels in prostatic adenocarcinoma as compared with normal prostatic tissues. Since its discovery, AMACR has gained wide acceptance for use in the diagnosis of prostatic adenocarcinoma in conjunction with morphology and immunohistochemical staining for basal cell markers. Numerous studies have consistently shown high sensitivity and specificity of AMACR for prostate cancer. This review focuses on AMACR expression in prostate cancer and its morphologic variants, high grade prostatic intraepithelial neoplasia, adenosis and benign conditions of the prostate. In addition, we discuss AMACR expression in other tumors. We also focus on the utility and technical aspects of the now-popular "triple stain" immunohistochemical antibody cocktail, consisting of antibodies to high-molecular-weight keratin, p63 and AMACR. Finally, we emphasize diagnostic pitfalls in the application of AMACR to small, atypical foci of glands seen on prostate needle core biopsy and project future diagnostic as well as clinical applications for the protein.     

Centrifugal separation of carcinoma or atypical cells in voided urine

A simple density gradient method was used to separate atypical and cancer cells from non-cancer cells in voided urine from patients with transitional cell atypia (moderate and grave atypia) and bladder cancer (squamous cell carcinoma and transitional cell carcinoma). Prior to cell separation, the Saccomanno preserved cells were dispersed by homogenization. After cell separation (5 min x 1400 rpm), atypical and cancer cells were enriched up to 20-fold. Also, most of the leucocytes (68-98%) and squamous cells (47-82%) were absent from density gradient specimen fractions containing the largest percentages of atypical and cancer cells. Peak purity ranges of atypical or cancer cells from different sample classes showed a large degree of overlap. This permitted the pooling of density gradient fractions enriched for atypical or cancer cells, thus increasing the efficiency of the method. Also, following centrifugation, the Papanicolaou-stained specimen fractions showed less background staining than the unprocessed controls, and the cells retained diagnostic morphologic features. We infer that this method may be a useful, low-cost approach for the morphologic study of developing cancers, not only from the urinary bladder, but also from the respiratory tract.     

Cytologic examination of post prostatic massage specimens as an aid in diagnosis of carcinoma of the prostate.

The results of cytologic examination fo post-massage prostatic secretion from symptomatic and asymptomatic subjects were evaluated by comparison with available tissue diagnosis. Specimens studied consisted of post-prostatic massage secretion and urine. The latter specimen gave much superior results. Although possibility of the cytologic diagnosis in prostatic carcinoma cases was limited, it proved to be reliable. Presently available short follow-up data indicate that this technique may play a role only as a diagnostic aid, however, its real value in prostatic cancer detection, could be only evaluated by long term follow-up of the "high risk" group.     

A digital image microscopy system for rare-event detection using fluorescent probes

Instrumentation for rare-event analysis should be capable of reliably detecting infrequent cells (less than 1:10,000) while both excluding false-positive signals and including true positive cells found in multicell clumps. We have developed a digital image microscopy (DIM) system in which a cytospin of 2 million cells is scanned with an intensified video camera (ISIT) using an IBM PC AT microcomputer-controlled microscope stage. PASCAL software controls the stage and analyzes video input, storing the location of positive cells to magnetic disk. The user can then "replay" each positive cell under computer control for either visual confirmation or analysis using other fluorescent probes. The computer requires 24 min to scan a cytoprep of 2 million cells, while playback for visual confirmation by the user averages 5 min. Using Hoechst-33342 premarked cells seeded into bone marrow as a model system, we found that the DIM system reliably detects one target cell per million marrow cells. With appropriate immunological markers, this system will aid in evaluating bone marrow purged of tumor cells prior to transplantation and should also be useful for detection of minimal residual disease in blood or bone marrow from patients with leukemia or solid tumors.     

Improved model for specimen classification based on single-cell classifiers

We consider probabilistic models for specimen classification procedures based on systems which classify individual cells as normal or abnormal. The models which we consider generalize those discussed previously by Castleman and White (Anal. Quant. Cytol. 2:117-122, 1980; Cytometry 2:155-158, 1981) and by Timmers and Gelsema (Cytometry 6:22-25, 1985). In particular, they include the biologically plausible possibility that the specimen contains cells which are intermediate between the extremes of normal and abnormal. We find that if these additional cells occur differentially in normal and abnormal specimens, then specimen classification can become substantially more efficient when the cell classifier has different error rates for these cells.     

Detection of bladder cancers using a SAMBA 200 cell image processor

The cell image analysis of urinary sediments was performed using a SAMBA 200 system. Cell profiles were created using 18 parameters related to size, shape, densitometry and chromatin texture. Learning sets of about 50 cell images per class were constructed for bening, degenerated benign, atypical, malignant and degenerated malignant urothelial cell types as well as for squamous epithelial and white blood cell types. A four-level hierarchic decision tree involving a discriminant analysis at each node was designed and then evaluated against a test set of 700 cells from the various classes. All of the cell images involved in this study were acquired from Papanicolaou-stained specimens obtained for routine screening. In spite of some misclassification errors, the analysis of the occurrence of cells in the various classes, especially the percentage of cells classified as suspicious (both atypical and malignant cells), by the SAMBA 200 system resulted in the separate clustering of the positive specimens (49 carcinomas grade II and higher) and the negative ones (26 benign samples). The preliminary results suggest that the cell population features (occurrence rate of cells in the various classes and mean cell profile within a class) may be of diagnostic value in designing a classifier dedicated to the prescreening of urinary sediments for the detection of bladder cancers.     

Interobserver and intraobserver differences in the diagnosis of urothelial cells. Comparison with classification by computer

The diagnostic performance of six human observers with various degrees of experience was tested against the diagnosis developed by consensus and by an automated, computer-generated cell-classification system. The study was based on randomly selected photographs of 200 urothelial cells, classified in groups of 50 as negative, atypical, suspicious and malignant, from smears of urine sediment. The performance of the human observers was tested in three separate sessions at suitable time intervals and evaluated by two analytical statistical techniques: the weighted kappa proposed by Cohen, to evaluate the degree of disagreement, and the log-linear analysis of association. The results indicated considerable interobserver and intraobserver differences, which appeared to be more closely related to the individual's perception of the visual targets than to the degree of experience or the target itself. The performance of the programmed computer was well within the midrange of the human diagnostic performance. This study indicated that under the experimental circumstances described in this paper, the diagnostic perception of cell images is highly subjective. The study implies that similar differences may exist elsewhere in diagnostic microscopy and suggests that the development of objective approaches to the morphologic classification of human disease may prove to be of clinical value.     

Cytologic phenomena accompanying uterine cervix electrocoagulation

In the study of cellular and tissue response to electrocoagulation of the uterine cervix, two cytologic phenomena accompanying the delayed healing process were described: the "contact-developed lucid cell" and the "regression field," which were limited to smear samples exhibiting the transitory appearance of abnormal cells after electrocoagulation. While "contact-developed lucid cells," which were firmly attached to abnormal target cell nuclei in a "cell-in-a-cell" pattern, had a variable effect upon subsequent smear scores, the "regression field," which is similar to that described by others during immunologic rejection of kidney transplants, was consistently followed by a shift of the smear score from the dysplasia range to the normal range. We suggest that abnormal cells differentiated during the healing process as well as those present both before and after treatment are subject to an immune rejection induced by uterine cervical electrocoagulation.     

Hemangiopericytoma in a male breast. Report of a case with cytologic, histologic and immunochemical studies

A hemangiopericytoma in a male breast was studied by fine needle aspiration (FNA) biopsy. The FNA smears contained tissue clumps showing knob-like formations of atypical cells, spindle-shaped cells and fragments of capillaries lined by normal endothelial cells. Immunocytochemical study showed a positive reaction for vimentin, but a negative reaction for desmin and keratin. Staining for Factor VIII was positive only in the capillaries and endothelial cells. The cytodiagnosis was "mesenchymal tumor." Histopathologic study of the mastectomy specimen made the final diagnosis of hemangiopericytoma. While FNA cytology and immunocytochemistry cannot make a definitive diagnosis of this rare vascular tumor, they can be decisive in planning the surgical treatment, as in the present case.     

Progression curves for endometrial lesions

OBJECTIVE: To derive a numeric measure for the progression of endometrial lesions as a baseline study for an eventual assessment of chemopreventive intervention efficacy. STUDY DESIGN: Tissue sections from normal endometrium at the proliferative and secretory phase, simple hyperplasia, atypical hyperplasia from cases free of concomitant adenocarcinoma and adenocarcinoma of the endometrium were recorded at high spatial resolution. Six cases from each diagnostic category were chosen as "typical," and 60 epithelial nuclei were randomly selected for measurement for each case. Discriminant analyses were carried out to derive a direction of progressive change in feature space and to correct the progression curve for the presence of cells not expressing progressive change among the random sample of nuclei. RESULTS: A well-conditioned progression curve was derived based on the mean discriminant function scores for each diagnostic category and the mean nuclear abnormality of the nuclei in each category, as expressed by their deviation in feature values from normal reference nuclei. The lesion signatures showed a clear trend toward extension into the range of higher nuclear abnormalities with increasing progression. There was an indication that abnormal endometrial lesions may comprise cases with distinctly different degrees of nuclear abnormality. CONCLUSION: A numeric assessment of lesion progression for endometrial lesions, based on karyometric measurements, is possible. The data suggest that additional analysis may provide further characterizing information for individual lesions.     

The use of hierarchic classification in the image analysis of a complex cell population. Experience with the sediment of voided urine

Microscopic analysis of cells in the sediment of voided urine is the principal noninvasive method of diagnosing and detecting cancer of the lower urinary tract, mainly the bladder. The sediments contain several populations of cells of unequal diagnostic value. By applying a system of hierarchic classification to the computer analysis of digitized cell images, we were able to eliminate from diagnostic consideration cells that are difficult to classify, such as degenerated cells, multinucleated cells, cell clusters, renal tubular cells and cells infected by the human polyomavirus. When this method of triage was applied to the images of sequentially encountered epithelial cells and clusters, the cell images accepted for final analysis by the computer were sufficient in number and quality to automatically construct cytologic profiles of documented diagnostic value in 15 patients with bladder cancer. The method proved to be applicable to smears and quantitative cytocentrifuge preparations processed by methods developed in our laboratory. This work clearly documents the feasibility of automated analysis of cells in voided urine for the purpose of diagnosing bladder cancer. It also confirms prior observations suggesting that a relatively small sample of sequential images of epithelial cells (200 to 300) is sufficient to establish a diagnostic profile of clinical value on patients with high-grade cancer of the urinary bladder.     

Fine needle aspiration cytology of pulmonary carcinoid tumors

Twenty-four cases coded as pulmonary carcinoid tumors initially sampled by fine needle aspiration (FNA) biopsy were reviewed in order to determine the cytologic features most useful in making the FNA diagnosis. The diagnosis of carcinoid tumor had been confirmed in 23 cases; the remaining case, though closely resembling a carcinoid tumor on the FNA specimen, proved to be a sclerosing hemangioma of the lung. Comparison of the original and review interpretations of the FNA specimens revealed that all typical spindle cell carcinoids and all atypical carcinoids were correctly diagnosed and classified. Of the 15 typical round cell carcinoids, the original cytologic diagnosis was lymphoma in 2 cases and benign bronchial lining cells in 2 cases. Thus, it appears that diagnostic errors are most likely in "typical" carcinoids. Review of the FNA findings suggests that the frequently stripped cytoplasm (with resulting non-cohesive bare nuclei), coupled with the almost universal plexiform vascularity (seen in 21 of 23 cases), should allow an accurate cytologic diagnosis in virtually all cases.     

Evaluation of the AutoCyte SCREEN system in a clinical cytopathology laboratory

OBJECTIVE: To compare the AutoCyte SCREEN (AutoCyte, Burlington, North Carolina, U.S.A.) system with manual screening by experienced cytotechnologists using thin-layer preparations that had been previously extensively studied and their cytologic abnormalities well defined. STUDY DESIGN: AutoCyte PREP (AutoCyte) samples prepared for a previous split-sample study comparing thin-layer preparations to conventional smears were used. These 1,992 AutoCyte PREP samples were in a cohort the abnormal findings of which had been confirmed via independent review by two sets of pathologists. For the current study, these samples were remasked and evaluated by the AutoCyte SCREEN system in a clinical laboratory. The instrument scanned each slide and selected six overview fields and 120 single objects for storage and display. The computer classified each slide in one of the following categories: abnormal, uncertain, normal or unsatisfactory. Independently for each case, a cytotechnologist evaluated the six fields and 120 objects selected by the instrument as abnormal, normal or unsatisfactory. For those cases classified as uncertain by AutoCyte, the technologist then reexamined the cellular displays and entered a consensus classification. These results were then compared to those of an independent review by cytotechnologists of the identical set of slides using routine manual screening. RESULTS: The AutoCyte SCREEN selected 35% of slides for manual review. Technologist and computer rendered equivalent classifications in 79%. Of the total slides screened by the AutoCyte SCREEN, 57% were classified as "uncertain," and 88% of these were subsequently classified as normal by consensus. Using the well-defined abnormal values of the cellular sample as a basis for calculation, the AutoCyte SCREEN-assisted practice had a diagnostic sensitivity of 85% and diagnostic specificity of 97.6%. Comparable values for manual screening of the identical cellular sample were a diagnostic sensitivity of 80% and specificity of 97.4%. CONCLUSION: The AutoCyte SCREEN achieves comparable or greater sensitivity in detecting cervical abnormalities in comparison with manual screening. When combined with the substantial advantage of thin-layer preparations over conventional smears, the AutoCyte SCREEN provides a screening system of superior sensitivity over conventionally prepared and examined cervical smears.     

First step toward the “fingerprinting” of brain tumors based on synchrotron radiation X-ray fluorescence and multiple discriminant analysis

Synchrotron-radiation-based X-ray fluorescence was applied to the elemental microimaging of neoplastic tissues in cases of various types of brain tumors. The following cases were studied: glioblastoma multiforme, gemistocytic astrocytoma, oligodendroglioma, anaplastic oligodendroglioma, ganglioglioma, fibrillary astrocytoma, and atypical transitional meningioma. Apart from neoplastic tissue, the analysis included areas of tissue apparently without malignant infiltration. The masses per unit area of P, S, Cl, K, Ca, Fe, Cu, Zn, Br, and Rb were used to construct a diagnostic classifier for brain tumors using multiple discriminant analysis. It was found that S, Cl, Cu, Fe, K, Br, and Zn are the most significant elements in the general discrimination of tumor type. The highest similarity in elemental composition was between atypical transitional meningioma and fibrillary astrocytoma. The smallest differentiation was between glioblastoma multiforme and oligodendroglioma. The mean percentage of correct classifications, estimated according to the a posteriori probabilities procedure, was 99.9%, whereas the mean prediction ability of 87.6% was achieved for ten new cases excluded previously from the model construction. The results showed that multiple discriminant analysis based on elemental composition of tissue may be a potentially valuable method assisting differentiation and/or classification of brain tumors.