Image analysis in anatomic pathology: what are the issues?

A classificação de Bethesda: uma análise morfométrica das biópsias do colo uterino

Com a possibilidade de se estender a classificação de Bethesda também para os laudos histológicos, consideramos neste trabalho alguns dos problemas mais importantes ligados às lesões relacionadas ao vírus da Papilomatose Humana (VPH), em particular a possibilidade de se estabelecer critérios para a definição diagnóstica entre lesões de alto e baixo grau. Comparamos características citológicas e histológicas de condições pré-cancerosas do colo uterino por meio de análise morfométrica. Cortes histológicos de 62 espécimes de biópsias cervicais guiadas por colposcópio foram estudadas. Os casos foram agrupados de acordo com o diagnóstico histopatológico em três grupos: normal (33 casos), Lesão lntraepitelial de baixo grau (LlE-BG, 21 casos) e Lesão Escamosa lntraepitelial de alto grau (LIE-AG, 8 casos). Foram criados dois programas DRAW e COUNT3, considerando os seguintes parâmetros: área, perímetro, diâmetro máximo e circularidade dos núcleos das células nas lesões, e a celularidade da camada superficial do epitélio cervical. Após as medições das células de interesse, os resultados eram submetidos a análise estatística multivariada, e computada a probabilidade posterior para confirmar a homogeneidade dos casos dentro do grupo diagnóstico. Area e celularidade parecem ser os parâmetros mais confiáveis. Apenas seis casos apresentaram a probabilidade posterior abaixo de 0,975. Os procedimentos morfométricos deram pleno suporte à distinção entre tecido escamoso normal, LlE-BG e LIE-AG.

On the concept of objectivity in digital image analysis in pathology

AIMS

The term 'objective' connotes a method that is based on facts and not influenced by personal opinions, perception or emotion. One often reads in the biomedical literature claims of objectivity for methods that use digital image analysis applied to histology. Since objective assessment of histology would represent a huge leap forward in scientific measurement and clinical diagnosis, such claims should be substantiated by strong evidence. This paper takes a selective look at the literature on image analysis to assess the definition of objectivity in image analysis and asks whether such a claim is ever justified.

METHODS

First, a brief background on the basic science of image analysis in histology details some of the controversies and opinions in the field. Then, a literature review of a subset of papers pertaining to image analysis in histology (with claims of objectivity) is conducted to determine what evidence exists for objectivity in these methods.

RESULTS

It was found that image analysis may have many benefits (speed, indefatigability, standardisation, etc.). However, algorithms are devised and implemented by human beings who make subjective decisions at each stage of the algorithm design and implementation process. Thus, image analysis methods can be seen as deterministic processes which 'objectively' implement the subjective decisions of the programmer. This indicates that 'inter-observer' variation in image analysis is equivalent to 'inter-algorithm' variation (which is rarely studied) and that a single computer algorithm's repeatability is of lesser importance than the repeatability of the image analysis method as a whole (including the block, slide and field selection and the method of tissue processing).

CONCLUSION

Repeatability and automaticity must not be confused with objectivity, but a lack of objectivity does not imply a lack of utility. Unless specific evidence of objectivity is provided, editors should insist that claims of objectivity in image analysis papers be either removed or justified prior to publication.

Evaluation of diagnostic efficiency of computerized image analysis based quantitative nuclear parameters in papillary and follicular thyroid tumors using paraffin-embedded tissue sections

Computerized image analysis (IA) system has emerged in recent years as a very powerful tool for objective and reproducible quantification of histological features. It has shown considerable potential for diagnostic application in diverse histological situations. The objectives of the present study were to evaluate the discriminatory diagnostic efficiency of computerized image analysis based quantitative subvisual nuclear parameters in papillary and follicular neoplasms of thyroid. A total of 60 cases were studied. Forty-four cases belonged to training set and 16 cases belonged to a test set. A minimum of 100 nuclei was analyzed in each case using uniform 5 m mm thick hematoxylin stained sections. The IA workstation comprised of an Olympus microscope, a 10 bit digital video camera, an image grabber card and a pentium 120 MHz computer. Optimas 5.2 software was utilized for data collection on 8 morphometric and 8 densitometric parameters. Multivariate stepwise discriminant statistical analysis of data was done with the help of BMDP statistical software release 7.0. Results from a training set revealed correct classification rates of 98.0%, 84.5% and 61.2% for the histological groups of hyperplastic papillae versus papillae of papillary carcinoma (group I), follicular variant of papillary carcinoma versus the broad category of follicular neoplasms consisting of both follicular adenoma and follicular carcinoma (group II) and follicular adenoma versus follicular carcinoma (group III), respectively. Results of test set revealed correct classification rates of 100%, 80% and 50% for groups I, II and III respectively. It was concluded that computerized nuclear IA parameters have potential usefulness for discriminating benign versus malignant papillary lesions of thyroid, follicular variant of papillary carcinoma versus follicular adenoma and/or follicular carcinoma but are of no value in discriminating between follicular adenoma and follicular carcinoma.

Morphometric differentiation between responsive tumor cells and mesothelial hyperplasia in second-look operations for ovarian cancer

We developed a procedure based on computerized image analysis to establish objective criteria for the differential diagnosis between mesothelial hyperplasia and cancer in peritoneal tissue samples obtained at second-look operations for ovarian cancer. The tumor tissue after chemotherapy was classified as "nonresponsive" if it was found by histologic criteria to be roughly similar to the tumor before chemotherapy and as "responsive" if it was found to be different (small clusters of bland-looking cells with no mitotic activity). Eighty-five samples of tissue had been classified previously by a pathologist into one of the four following groups: ovarian tumor prior to chemotherapy, "responsive" tumor, "nonresponsive" tumor, or mesothelial hyperplasia. Cell profiles of the tissue samples were studied by computerized image analysis using 21 morphometric descriptors derived from the manual tracings of tumor nuclei, including nuclear perimeter, nuclear area, maximal chord, circularity factor, and standard deviations of these descriptors. Size distribution curves of nuclear areas and maximal chords were included in the analysis. A multivariate discriminant analysis confirmed the separation into the four diagnostic groups, accomplished with consideration of the physical descriptors alone, except for some overlapping between groups 1 and 3. The separation between carcinoma and mesothelial hyperplasia was clear in all cases.

The microcomputer and image analysis in diagnostic pathology

This paper presents a snapshot view of the influence and direction of microcomputer technology for image analysis techniques in diagnostic pathology. Microcomputers have had considerable impact in bringing image analysis to wider application. Semi-automated tracing techniques are a simple means of providing objective data and assist in a wide range of diagnostic problems. From the common theme of reducing subjectivity in diagnostic assessment, an extensive body of research has accrued. Some studies have addressed the need for quality control for reliable, routine application. Video digitizer cards bring digital image analysis within the reach of laboratory budgets, providing powerful tools for investigation of a wide range of cellular and tissue features. The use of staining procedures compatible with quantitative evaluation has become equally important. As well as assisting scene segmentation, cytochemical and immunochemical staining techniques relate the data to biological processes. With the present state of the art, practical use of microcomputer based image analysis is impaired by limitations of information extraction and specimen throughput. Recent advances in colour video imaging provide an extra dimension in the analysis of multi-spectral stains. Improvements will also be felt with predictable increase in speed of microprocessors, and with single chip devices which deliver video rate processing. If the full potential of this hardware is realized, high-speed, routine analysis becomes feasible. In addition, a microcomputer imaging system can play host to companion functions, such as image archiving and transmission. With this outlook, the use of microcomputers for image analysis in diagnostic pathology is certain to increase.(ABSTRACT TRUNCATED AT 250 WORDS)

Observer variation in quantification of immunocytochemistry by image analysis

This paper reports the findings of a study designed to examine observer variation as a source of inaccuracy inherent in the use of computer-assisted image analysis to measure areas of stained tissue. The rat pituitary immunostained for prolactin and galanin was used as an example to estimate patterns of immunoreactivity exhibited by different cell types. Six observers, with differing experience, selected grey level threshold values on 40 fields of images of stained tissue making three repeats of each field. The 40 fields consisted of 20 serial pairs of colocalized fields, one immunostained for prolactin, the other for galanin. The 20 pairs consisted of four pairs from each of five animals. Analysis of observer variation in the selection of threshold values showed large differences in the within- and between-observer variation. Analysis of the components of variance in the estimation of the ratios of stained tissues showed that the major source of variation was the within-observer component. An additional experiment using two observers, where half of the images were compared to the original microscope images before setting threshold levels, showed that the opportunity to make a comparison did not reduce observer variation. It is suggested that any study which uses semi-automatic methods to segment regions of a digital image can benefit from an analysis of this kind so that the sources of variation can be determined to enable maximum discriminating power in future studies.

Ultrastructural discrimination between malignant melanomas and benign nevocytic nevi using high-resolution image and multivariate analyses

Prompted by the well-known difficulties of reliable and objective histologic differentiation between initial malignant melanoma (MM) and benign nevocytic nevi (NN), ultrastructural high-resolution image and multivariate analyses were evaluated for their diagnostic efficiency. Thirty-seven different features describing morphometry (area, circumference, and shape factor), amount of heterochromatin and euchromatin, chromatin homogeneity, and presence of smaller dark chromatin aggregations were determined by a MICROVAX 3500 computer in each of 1840 intraepidermal melanocytic nuclei of 17 MM and 20 NN. A strategy for the classification of cases based on the identification of markedly atypical melanocytic cells (MACS) was developed. MACS, selected in multivariate analysis with a linear combination of the eight most important features for cell classification, were found in 39.4% of the melanoma cells, but only in 0.3% of nevocytic nevus cells. The presence of MACS allowed a clear differentiation between MM and NN. All cases of MM had more than four MACS, whereas 17 cases of nevocytic nevi were MACS negative, and in each of the remaining three cases only one MAC was present. The percentage of MACS detected within intraepidermal parts of MM by using computerized high-resolution image analysis was found to be a highly efficient diagnostic marker. The new classification strategy has the potential of saving considerable time in subsequent studies, because preselected sampling and the calculation of only a few criteria have proven sufficient for correct classification of malignant melanomas.

How malignant is malignant? A brief review of the microscopic assessment of human neoplasms, and the prediction of whether they will metastasize and kill

This review examines ways in which histopathologists can attempt to predict the degree of malignancy from the microscopic examination of surgically removed samples of human neoplasms. These include mitotic counts, measurement of DNA content, analysis of DNA turnover, measurement of proliferation antigens, measurement of nucleolar organizing areas, counting nucleoli, morphometry of nuclei, identification of genetic change, measurement of lysis of basement membrane, and measurement of invasion in vitro. The latter two are not currently of practical value. Of the rest none is uniformly capable of predicting the degree of malignancy of a human neoplasm. After further refinement and control measurement of several such parameters may however contribute more to the prediction of malignancy than present subjective assessments.

Quantitative nuclear cell image analyses of thyroid tumors from archival material

Sixty-three sections of Feulgen-stained thyroid cell nuclei from paraffin-embedded material, including five multinodular goiters, 10 adenomas, 36 papillary carcinomas, seven follicular carcinomas, and five medullary carcinomas were analyzed by means of the SAMBA 200 (TITN, Grenoble, France) cell image processor. This was done in order to obtain nuclear characteristics of papillary versus follicular carcinomas. The nuclear features were assessed by morphometric, densitometric, and textural parameters. Our preliminary results indicate that the cell nuclei from typical histopathologic specimens of follicular thyroid cancers belong to a larger thyroid cell nuclei population corresponding to the histopathologic family of papillary thyroid cancers. This follicular neoplastic cell nuclei population appears to be quite distinct from the typical medullary neoplastic cell nuclei population which also belongs to the papillary neoplastic cell nuclei population. It appears that there is a specific papillary cell nuclei subpopulation containing typical hypochromatic cell nuclei. We also observed a dramatic increase in nuclear size and hyperchromatism between normal (multinodular goiters) and neoplastic (carcinomas) thyroid tissues, with the benign tissues (adenomas) showing intermediate nuclear characteristics.

Morphometric analysis of fine needle aspirates from breast lesions

In an attempt to discover the morphometric variables with the most diagnostic power in the differentiation of benign from malignant breast disease, 20 unequivocally benign and 20 unequivocally malignant and histologically confirmed breast aspirates were examined on an image analyser. It was found that standard deviation of nuclear area was the most discriminant variable. Then 23 aspirates initially diagnosed as 'suspicious of malignancy' were measured by the same technique, and standard deviation of nuclear area correctly differentiated all but three cases.

Interactive morphometry of normal and hyperplastic peritoneal mesothelial cells and dysplastic and malignant ovarian cells

This study used computerized interactive morphometry to evaluate the differential characteristics of mesothelial normal cells, mesothelial hyperplastic cells, and carcinomatous cells, and also compared hyperplastic mesothelial cells with ovarian dysplastic cells from a previous study. The procedure included extraction of multiple descriptors of the nuclear profile: perimeter length, area, longest chord, circularity factors, standard deviations of these characteristics, and a 10-bin size distribution table of the nuclear area. The final classification is achieved by stepwise discriminant analysis of these variables. The analysis classified all cases correctly with high posterior probabilities.

Influence of smear preparation and fixatives on the DNA ploidy and the morphonuclear features of the MXT mammary tumor and normal tissues in the mouse

We compared cytomorphonuclear parameters--related to DNA histogram and chromatin distribution--on MXT mouse mammary tumor and murine normal cells from fresh squash smears or from deparaffinized tissue smears fixed in several fluids. We used the SAMBA 200 cell image processor with software allowing for the discrimination of parameters computed on Feulgen-stained nuclei. The spectrophotometric results--assessed by integrated optical density values--indicate that the nuclei from deparaffinized tissue fixed in Bouin's fluid are around 50% less stained than those fixed in formalin or ethanol-formalin-acetic acid (EFA). The fresh smears of nuclei fixed in formalin contain a less well-defined and more homogeneous chromatin than after Bouin's or EFA fixation. This has led to the conclusion that morphonuclear parameter comparisons performed on tissues differently processed or from different origins present severe limitations.

Interactive morphometric procedures and statistical analysis in the diagnosis of ovarian dysplasia and carcinoma

We report on our continued experience with an interactive morphometric method recently introduced by us for the definition and diagnosis of ovarian dysplasia vs. normal or malignant epithelium. The main quantitative differences between these three diagnostic categories are based on 1) cytology of the nuclei (nuclear area, circularity factor, maximum chord) and 2) on stratification (distances of nuclear centers to the basement membrane and number of cells per unit length of basement membrane). We implemented our approach on live video images viewed on a monitor overlaid with a touch sensitive screen by one of two interactive procedures: 1) by tracing nuclear profiles (procedure DRAW) or 2) by tracing the basement membrane and touching the center of all nuclei (procedure NU-MEAS). In all cases statistical analysis was performed on a string of multiple variables by stepwise discriminant analysis. Now we have straightened our data basis and are able to obtain diagnosis of unknown samples with very high posterior probabilities. Both procedures are effective but NU-MEAS requires the least effort and seems to give the best statistics.