Causal probabilistic modeling for malignancy grading in pathology with explanations of dependency to the related histological features

This work demonstrates that histological grading of brain tumors and astrocytomas can be accurately predicted and causally explained with the help of causal probabilistic models, also known as Bayesian networks (BN). Although created statistically, this allows individual identification of the grade of malignancy as an internal cause that has enabled the development of the histological features to their observed state. The BN models are built from data representing 794 cases of astrocytomas with their malignant grading and corresponding histological features. The computerized learning process is improved when pre-specified knowledge (from the pathologist) about simple dependency relations to the histological features is taken into account. We use the BN models for both grading and causal analysis. In addition, the BN models provide a causal explanation of dependency between the histological features and the grading. This can offer the biggest potential for choice of an efficient treatment, since it concentrates on the malignancy grade as the cause of pathological observations. The causal analysis shows that all ten histological features are important for the grading. The histological features are causally ordered, implying that features of first order are of higher priority, e.g. for the choice of treatment in order not to allow the malignancy to progress to a higher degree. Due to the explanations of feature relations, the causal analysis can be considered as a powerful complement to any malignancy classification tool and allows reproducible comparison of malignancy grading.     

A modern reproducible method for the histologic grading of astrocytomas with statistical classification tools

OBJECTIVE: To investigate whether statistical classification tools can infer the correct World Health Organization (WHO) grade from standardized histologic features in astrocytomas and how these tools compare with GRADO-IGL, an earlier computer-assisted method. STUDY DESIGN: A total of 794 human brain astrocytomas were studied between January 1976 and June 2005. The presence of 50 histologic features was rated in 4 categories from 0 (not present) to 3 (abundant) by visual inspection of the sections under a microscope. All tumors were also classified with the corresponding WHO grade between I and IV. We tested the prediction performance of several statistical classification tools (learning vector quantization [LVQ], supervised relevance neural gas [SRNG], support vector machines [SVM], and generalized regression neural network [GRNN]) for this data set. RESULTS: The WHO grade was predicted correctly from histologic features in close to 80% of the cases by 2 modern classifiers (SRNG and SVM), and GRADO-IGL was predicted correctly in > 84% of the cases by a GRNN. CONCLUSION: A standardized report, based the 50 histologic features, can be used in conjunction with modern classification tools as an objective and reproducible method for histologic grading of astrocytomas.     

Cytological grading of breast cancer in Giemsa-stained fine needle aspiration smears

A potential cytological nuclear grading based on a semi-quantitative evaluation of three basic nuclear features, size of cell nuclei, anisonucleosis and the proportion of nucleoli-containing-nuclei, was tested on 74 Giemsa-stained fine needle aspiration of breast smears for its reliability in establishing the malignant potential of breast cancer. The prognostic impact of DNA-ploidy and S-phase fraction was also assessed. A good correlation between the three basic nuclear features, DNA-ploidy, S-phase fraction, cytological nuclear grade and histological grade, was shown. Using the cytological nuclear grade proposed, correct classification of cases between low histological grade (HG I) and high histological grade (HG II + HG III) was achieved in 79.73%. A statistically significant difference in 5-year survival rate was also observed between low malignancy grade and high malignancy grade breast cancer patients, regardless of the grading method used. DNA-ploidy and S-phase fraction were not statistically significant in establishing the malignant potential of breast cancer.     

Neural network-based segmentation and classification system for automated grading of histologic sections of bladder carcinoma

OBJECTIVE: To develop an image analysis system for automated nuclear segmentation and classification of histologic bladder sections employing quantitative nuclear features. STUDY DESIGN: Ninety-two cases were classified into three classes by experienced pathologists according to the WHO grading system: 18 cases as grade 1, 45 as grade 2, and 29 as grade 3. Nuclear segmentation was performed by means of an artificial neural network (ANN)-based pixel classification algorithm, and each case was represented by 36 nuclei features. Automated grading of bladder tumor histologic sections was performed by an ANN classifier implemented in a two-stage hierarchic tree. RESULTS: On average, 95% of the nuclei were correctly detected. At the first stage of the hierarchic tree, classifier performance in discriminating between cases of grade 1 and 2 and cases of grade 3 was 89%. At the second stage, 79% of grade 1 cases were correctly distinguished from grade 2 cases. CONCLUSION: The proposed image analysis system provides the means to reduce subjectivity in grading bladder tumors and may contribute to more accurate diagnosis and prognosis since it relies on nuclear features, the value of which has been confirmed.     

Karyometric features of low and high grade glial tumors as compared with their MRI appearance

OBJECTIVE: To compare the results of a magnetic resonance imaging (MRI) grading designed to identify low and high grade gliomas with karyometry used as a tool to grade primary brain tumors. STUDY DESIGN: A consecutive series of 23 primary brain tumors was selected for this study. The neuroradiologist, not knowing the histologic diagnoses, divided the cases into low and high grade categories on the basis of the following 7 features: border sharpness, heterogeneity without contrast, cavitation, contrast enhancement, hypervascularity, mass effect and perifocal T2 hyperintensity. To each feature was given a numerical value, ranging from 1 to 3. All the cases were reviewed and classified by the same pathologist, blinded to the MRI diagnosis. Two hundred nuclei per case were recorded, and 93 karyometric features related to nuclear area, total optical density and chromatin distribution were analyzed for each nucleus. Statistical analysis included discriminant analysis, Kruskal-Wallis test, nonsupervised learning algorithm P-index and Beale statistic. RESULTS: Ten cases were classified as low grade on the basis of their MRI features. The corresponding histopathologic diagnoses were: grade 2 astrocytoma in 2 cases and grade 2 oligodendroglioma in 8 cases. An MRI diagnosis of high grade tumor was made in 13 cases. In 10 cases it was confirmed by the histopathologic diagnosis (3 grade 3 astrocytomas, 1 grade 3 oligodendroglioma and 6 glioblastomas). In the remaining 3 cases the histologic examination revealed a low grade tumor, 1 grade 2 astrocytoma and 2 grade 2 oligodendrogliomas. For the purposes of the karyometric analysis the cases were allocated to the low or high grade category according to their histologic diagnosis (13 cases low grade and 10 cases high grade). Nuclei from low and high grade tumors showed clearly different karyometric characteristics. The oligodendroglioma nuclei had abnormality values close to the low grade standard, while the astrocytoma nuclei were a highly dispersed group with characteristics indicative of a higher degree of nuclear abnormality than the oligodendroglioma nuclei. The results of karyometric analysis showed that grade 2 tumors, corresponding to the low grade group, form a rather distinct category from grade 3 and 4 tumors belonging to the high grade group. CONCLUSION: The results of MRI grading based on a series of features that are routinely assessed by the neuroradiologist to reach a final diagnosis correlate highly with the histopathologic diagnosis. Karyometry can be a useful adjunct to histologic grading.     

Computer-assisted grading of adenocarcinoma in prostatic aspirates

Conventional cytologic grading of fine needle aspirates of prostatic adenocarcinoma has been shown neither to be reproducible nor to correlate well with histologic grading. This study developed a tumor grade classification based on computerized cytomorphometric features and compared the results to conventional grading of companion tissue sections. The image analysis system evaluated architectural features of the aspirates (mainly cell cluster features and interrelationships) as well as nuclear features. Thirty-five prostatic adenocarcinomas (8 well, 19 moderately and 8 poorly differentiated) were evaluated. Discriminant functions based on data collected at medium and high resolution distinguished between aspirates from low-grade (well-differentiated) and high-grade (poorly differentiated) adenocarcinomas with 81% accuracy. Moderately differentiated cancers could not be classified as a distinct group. This study suggests that accurate grading of prostatic adenocarcinoma in fine needle aspirate smears requires the evaluation of medium-resolution features related to specimen cellularity and uniformity or crowding of cell clusters as well as of high-resolution features of nuclear area, perimeter and coarseness of chromatin texture. These findings are compared to those of other schemes for the cytologic grading of prostatic aspirates.     

Automated diagnosis of brain tumours astrocytomas using probabilistic neural network clustering and support vector machines

A computer-aided diagnosis system was developed for assisting brain astrocytomas malignancy grading. Microscopy images from 140 astrocytic biopsies were digitized and cell nuclei were automatically segmented using a Probabilistic Neural Network pixel-based clustering algorithm. A decision tree classification scheme was constructed to discriminate low, intermediate and high-grade tumours by analyzing nuclear features extracted from segmented nuclei with a Support Vector Machine classifier. Nuclei were segmented with an average accuracy of 86.5%. Low, intermediate, and high-grade tumours were identified with 95%, 88.3%, and 91% accuracies respectively. The proposed algorithm could be used as a second opinion tool for the histopathologists.     

A comparative study in morphometric grading of transitional cell carcinoma of the urinary bladder

To overcome the considerable observer inconsistency in the histologic grading of transitional cell carcinomas, the value of four different morphometric grading methods was investigated in 61 tumors of the bladder. Only two methods showed satisfactory reproducibility. Both methods, one based on random nuclear sampling and the other on selective nuclear sampling, showed an increase in the mean and standard deviation of the nuclear area with higher tumor grades (P less than .00001). Morphometric classification of the learning set (44 cases) was in agreement with the unequivocally assessed histologic grade in 35 cases (79.5%) using random sampling and in 38 cases (86.4%) using selective sampling. By reducing the grading classes to "low" (grades 1 and 2) and "high" (grade 3) and by introducing a classification probability threshold (0.80), an accurate morphometric classification was achieved in 38 cases (86.4%) using random sampling and in 41 cases (93.2%) using selective sampling. Of the 17 cases with histologic grading discrepancies, all 10 low-grade tumors (with discrepancies of grade 1 versus grade 2) were correctly classified as low-grade carcinomas by both of the morphometric methods; in the remaining 7 cases, with low-versus-high discrepancies (grade 2 versus grade 3), the selective method yielded better correlation with the tumor stage and clinical follow-up. It is concluded that morphometric classification is an acceptable alternative for histologic grading by pathologists, provided that the reproducibility of the method is confirmed. Although both random and selective sampling yielded satisfactory classifications, the selective method gave more reliable results as confirmed by the clinical behavior.     

Further testing of morphometric grading in T(A,1) urothelial carcinomas of the urinary bladder and the additional value of p53 immunoquantitation

OBJECTIVE: To analyze the value for grading of a previously developed quantitative morphometric/cytometric multivariate grading model (consisting of the mean nuclear area of the 10 largest nuclei (MNA-10, mitotic activity index = MAI and Ki-67 area% = Ki-67) in two new independent test sets of urothelial carcinomas (UCs) of the urinary bladder and to evaluate the additional value of p53 area% (p53) in this model. STUDY DESIGN: Ki-67 immunoquantitation, morphometric MAI and MNA-10 assessments using a previously described, strict protocol and matching of the resulting morphometric grade with subjective grade in two test sets of 154 T(A,1) UCs of the bladder (consensus grade between two independent observers). Further testing of this morphometric grading model was performed in 57 cases that lacked initial interobserver agreement on grade. Single and multivariate analysis of all features (including p53) was performed. RESULTS: With the previously developed morphometric/cytometric grading model, 93% (grade 1 vs. 2) and 91% (grade 2 vs. 3) of the consensus cases were correctly classified. These percentages were very similar to previous results, suggesting that the model is robust. Of the 57 cases that lacked initial interobserver agreement on grade, 53/57 (93%) were classified unambiguously as grade 1, 2 or 3 with the quantitative morphometric/ cytometric grading model. In the exploratory analysis, p53 was significant but with more overlap than the other features had. In multivariate analysis p53 did not improve the overall classification result of the original morphometric/cytometric model. CONCLUSION: The value of MNA-10, MAI and Ki-67 for grading in T(A,1) urothelial carcinomas of the urinary bladder was confirmed. p53 Did not improve overall grading classification of this combination.     

Histometric features for the grading of prostatic carcinoma

Histometric features for the objective grading of prostatic adenocarcinoma in histologic specimens were analyzed in five cases each of well, moderately and poorly differentiated lesions. Tissue sections from the selected cases were stained by the Feulgen method and digitized by a video-based microphotometer. Twenty total fields were recorded for each grade: ten at high resolution (an image sampling of 0.5 micron per pixel) and ten at low resolution (0.8 micron per pixel), with two fields per case recorded at each resolution. The images were segmented by an automated expert system-guided scene segmentation procedure. The performance of that procedure was measured by comparing the automated counts of nuclei in the segmented fields to the visual counts made by a pathologist in the same fields. For well, moderately and poorly differentiated cases, respectively, the nuclear counts made by the expert system at high resolution were 2.7%, 4.2% and 4.7% higher than the visual counts (as estimated from a total of 6,628 nuclei), but 1.2%, 2.5% and 1.1% lower at low resolution (10,329 nuclei). High-resolution features and tissue textural features were computed for each case. The high-resolution features showed good separation between the three groups of cases. The tissue textural features showed consistent separation between well and moderately differentiated cases. The relaxation of the spatial resolution (to 0.8 micron/pixel spacing) did not affect the selection of features, but led to less separation between the data from different grades. In conclusion, the automated system performed satisfactorily in distinguishing sections of prostatic tumors of varying degrees of differentiation.     

Grade classification of neuroepithelial tumors using high-resolution magic-angle spinning proton nuclear magnetic resonance spectroscopy and pattern recognition

Clinical data have shown that survival rates vary considerably among brain tumor patients, according to the type and grade of the tumor. Metabolite profiles of intact tumor tissues measured with high-resolution magic-angle spinning proton nuclear magnetic resonance spectroscopy (HRMAS (1)H NMRS) can provide important information on tumor biology and metabolism. These metabolic fingerprints can then be used for tumor classification and grading, with great potential value for tumor diagnosis. We studied the metabolic characteristics of 30 neuroepithelial tumor biopsies, including two astrocytomas (grade I), 12 astrocytomas (grade II), eight anaplastic astrocytomas (grade III), three glioblastomas (grade IV) and five medulloblastomas (grade IV) from 30 patients using HRMAS (1)H NMRS. The results were correlated with pathological features using multivariate data analysis, including principal component analysis (PCA). There were significant differences in the levels of N-acetyl-aspartate (NAA), creatine, myo-inositol, glycine and lactate between tumors of different grades (P<0.05). There were also significant differences in the ratios of NAA/creatine, lactate/creatine, myo-inositol/creatine, glycine/creatine, scyllo-inositol/creatine and alanine/creatine (P<0.05). A soft independent modeling of class analogy model produced a predictive accuracy of 87% for high-grade (grade III-IV) brain tumors with a sensitivity of 87% and a specificity of 93%. HRMAS (1)H NMR spectroscopy in conjunction with pattern recognition thus provides a potentially useful tool for the rapid and accurate classification of human brain tumor grades.     

DNA grading of malignancy in breast cancer. Prognostic validity, reproducibility and comparison with other classifications

The prognostic significance of the "DNA malignancy grade" (DNA-MG) was tested in a series of 104 breast cancer patients in comparison with TNM staging, histomorphologic grading according to Bloom and Richardson, mean nuclear area (MNA) and DNA-histogram classification according to Auer. The reproducibility and representativity of the grading systems were investigated, and their results in primary tumors and lymph node metastases were compared. The scalar DNA-MG was assessed on monolayer smears prepared from paraffin-embedded tissues; the smears were automatically Feulgen stained and used for rapid interactive DNA cytometric evaluation by an automated microscope and a TV image-analysis system. TNM staging showed the highest correlation with survival, followed by histomorphologic grading and DNA-MG; MNA and the DNA-histogram classification failed to give statistically significant prognostic information. Both histomorphologic grading and DNA-MG were identified as parameters adding independent prognostic information to the TNM staging. However, only DNA-MG demonstrated an acceptable reliability, with small 95% ranges between repeated measurements within the primary tumor (+/- 0.3 DNA-MG) and a strong correlation between the results in the primary tumor and its lymph node metastases. These findings show that the DNA-MG is a valid and reliable prognostic index that adds significant prognostic information to TNM staging.     

Distinguishing and grading human gliomas by IR spectroscopy

As a molecular probe of tissue composition, IR spectroscopy can potentially serve as an adjunct to histopathology in detecting and diagnosing disease. This study demonstrates that cancerous brain tissue (astrocytoma, glioblastoma) is distinguishable from control tissue on the basis of the IR spectra of thin tissue sections. It is further shown that the IR spectra of astrocytoma and glioblastoma affected tissue can be discriminated from one another, thus providing insight into the malignancy grade of the tissue. Both the spectra and the methods employed for their classification reveal characteristic differences in tissue composition. In particular, the nature and relative amounts of brain lipids, including both the gangliosides and phospholipids, appear to be altered in cancerous compared to control tissue. Using a genetic classification approach, classification success rates of up to 89% accuracy were obtained, depending on the number of regions included in the model. The diagnostic potential and practical applications of IR spectroscopy in brain tumor diagnosis are discussed.     

Computer-aided classification of malignancy in astrocytomas. I. The value of nuclear parameters obtained by automated black and white image analysis.

Ninety-three selected cases of astrocytomas including glioblastomas (astrocytomas grades 1-4) were evaluated by means of Feulgen-stained microscopic slides for nuclear parameters obtained by automated black and white image analysis (ABWIA). The goal was to determine to what extent nuclear features evaluated by ABWIA were applicable as classifiers for the computer-aided numerical classification of malignancy in astrocytomas. Before the automated evaluation, all tumours had been subjectively graded according to the Mayo Clinic grading rules as delineated by Ringertz. Twenty-three nuclear parameters were evaluated and tested for their classification impact. With a model of five parameters (number of nuclei per area, mean of the convex form factor, extinction sum, extinction variation, and full-width-half-maximum of the extinction distribution) the highest reclassification rate of 75% correctly reclassified cases was obtained. Although this is a good result for a classification using only nuclear parameters, it is too poor for practical application. Thus, nuclear parameters evaluated by ABWIA alone are insufficient for numerical classification models assessing the malignant expression of astrocytomas.     

Cytologic grading of invasive breast carcinoma. Correlation with clinicopathologic variables and predictive value of nodal metastasis

OBJECTIVE: To estimate cytologic grade and correlate it with the other known prognostic factors, such as tumor differentiation, growth fraction, estrogen receptor status and nodal status. STUDY DESIGN: Fine needle aspirates from 104 invasive ductal carcinomas were stained by the Papanicolaou method and examined for necrosis, cellular size, nuclear/cytoplasmic ratio, nuclear pleomorphism, nucleoli, chromatin granularity and density of chromatin. We established a semiquantitative scoring system based on the above features and correlated cytologic findings with clinicopathologic variables. RESULTS: Histologic grade correlated positively with cytologic grade and negatively with estrogen receptor positivity. Moreover, high cytologic grade was associated with nodal metastasis and proliferative index labeling by MIB-1. CONCLUSION: This study showed that our grading system for breast cancer on fine needle aspiration cytology is feasible on a routine diagnostic basis. Cytologic grading can provide more information than usual on tumor biologic behavior.     

Nuclear texture features for classifying benign vs. dysplastic or malignant squamous epithelium of the larynx.

OBJECTIVE: To search for nuclear features and feature combinations able to assess malignancy and premalignant changes on tissue sections of laryngeal squamous epithelium. STUDY DESIGN: A total of 139 lesions of benign changes (BC) (n = 44), epithelial dysplasias (ED) (n = 50) and invasive laryngeal cancer (LC) (n = 45) were retrieved from archival pathology specimens. The goal of this study was to identify the best features or feature combinations that discriminate BC from LC and also reflect the degree of ED. In order to verify the results on independent data, the groups were split into two separate subgroups, one for training and one for testing. RESULTS: On the test set of slides, the overall correct classification of BC vs. LC cases was 82% using only one feature, fractal2_area. This classification rate could be increased to 91% when a discriminant function based on 10 features was used. However, this gain was not significant. CONCLUSION: Fractal texture features can be used to assess malignancy on tissue sections as an alternative to DNA measurement. In this study feature combinations did not significantly improve classification rates.     

Pathologic significance of proliferative activity and oncoprotein expression in astrocytic tumors

OBJECTIVE: To clarify the clinicopathologic significance of immunohistochemistry for proliferative activity and oncoprotein expression in astrocytic tumors. STUDY DESIGN: Ninety-seven cases of brain astrocytic tumors with histologic grading and follow-up data were investigated with immunohistochemistry and image analyzer to detect the expression of proliferating cell nuclear antigen (PCNA), silver-binding nucleolar organizer regions (AgNORs) and several oncoproteins. RESULTS: PCNA was significantly related to AgNORs, grading and prognosis of astrocytomas. The frequency of mutant p53 protein expression was higher in grade 2-4 astrocytomas than in grade 1. Epidermal growth factor (EGF) (37.1%), EGF receptor (83.5%) and p21ras (42.3%) expression levels were related to neither the grade nor prognosis of the tumors. The positive ratios of p53 antibodies were higher in grades 2-4 than in grade 1, and the intensities correlated with PCNA but not with prognosis. CONCLUSION: Aberrations of c-erbB-2, p21ras, EGF and EGF receptor might be early events in the initiation and progression of astrocytomas, whereas p53 overexpression is involved in all the stages. Immunohistochemical detection had no prognostic value. PCNA could be important to the evaluation of astrocytoma malignancy.     

MRI-based radiomics signature for tumor grading of rectal carcinoma using random forest model

The present study aimed to construct prospective models for tumor grading of rectal carcinoma by using magnetic resonance (MR)-based radiomics features. A set of 118 patients with rectal carcinoma was analyzed. After imbalance-adjustments of the data using Synthetic Minority Oversampling Technique (SMOTE), the final data set was randomized into the training set and validation set at the ratio of 3:1. The radiomics features were captured from manually segmented lesion of magnetic resonance imaging (MRI). The most related radiomics features were selected using the random forest model by calculating the Gini importance of initial extracted characteristics. A random forest classifier model was constructed using the top important features. The classifier model performance was evaluated via receive operator characteristic curve and area under the curve (AUC). A total of 1,131 radiomics features were extracted from segmented lesion. The top 50 most important features were selected to construct a random forest classifier model. The AUC values of grade 1, 2, 3, and 4 for training set were 0.918, 0.822, 0.775, and 1.000, respectively, and the corresponding AUC values for testing set were 0.717, 0.683, 0.690, and 0.827 separately. The developed feature selection method and machine learning-based prediction models using radiomics features of MRI show a relatively acceptable performance in tumor grading of rectal carcinoma and could distinguish the tumor subjects from the healthy ones, which is important for the prognosis of cancer patients.     

Noninvasive Characterization of Locally Advanced Breast Cancer Using Textural Analysis of Quantitative Ultrasound Parametric Images

PURPOSE: The identification of tumor pathologic characteristics is an important part of breast cancer diagnosis, prognosis, and treatment planning but currently requires biopsy as its standard. Here, we investigated a noninvasive quantitative ultrasound method for the characterization of breast tumors in terms of their histologic grade, which can be used with clinical diagnostic ultrasound data. METHODS: Tumors of 57 locally advanced breast cancer patients were analyzed as part of this study. Seven quantitative ultrasound parameters were determined from each tumor region from the radiofrequency data, including mid-band fit, spectral slope, 0-MHz intercept, scatterer spacing, attenuation coefficient estimate, average scatterer diameter, and average acoustic concentration. Parametric maps were generated corresponding to the region of interest, from which four textural features, including contrast, energy, homogeneity, and correlation, were determined as further tumor characterization parameters. Data were examined on the basis of tumor subtypes based on histologic grade (grade I versus grade II to III). RESULTS: Linear discriminant analysis of the means of the parametric maps resulted in classification accuracy of 79%. On the other hand, the linear combination of the texture features of the parametric maps resulted in classification accuracy of 82%. Finally, when both the means and textures of the parametric maps were combined, the best classification accuracy was obtained (86%). CONCLUSIONS: Textural characteristics of quantitative ultrasound spectral parametric maps provided discriminant information about different types of breast tumors. The use of texture features significantly improved the results of ultrasonic tumor characterization compared to conventional mean values. Thus, this study suggests that texture-based quantitative ultrasound analysis of in vivo breast tumors can provide complementary diagnostic information about tumor histologic characteristics.