Automatic grading of prostate cancer in digitized histopathology images: Learning from multiple experts.

Prostate cancer (PCa) is a heterogeneous disease that is manifested in a diverse range of histologic patterns and its grading is therefore associated with an inter-observer variability among pathologists, which may lead to an under- or over-treatment of patients. In this work, we develop a computer aided diagnosis system for automatic grading of PCa in digitized histopathology images using supervised learning methods. Our pipeline comprises extraction of multi-scale features that include glandular, cellular, and image-based features. A number of novel features are proposed based on intra- and inter-nuclei properties; these features are shown to be among the most important ones for classification. We train our classifiers on 333 tissue microarray (TMA) cores that were sampled from 231 radical prostatectomy patients and annotated in detail by six pathologists for different Gleason grades. We also demonstrate the TMA-trained classifier's performance on additional 230 whole-mount slides of 56 patients, independent of the training dataset, by examining the automatic grading on manually marked lesions and randomly sampled 10% of the benign tissue. For the first time, we incorporate a probabilistic approach for supervised learning by multiple experts to account for the inter-observer grading variability. Through cross-validation experiments, the overall grading agreement of the classifier with the pathologists was found to be an unweighted kappa of 0.51, while the overall agreements between each pathologist and the others ranged from 0.45 to 0.62. These results suggest that our classifier's performance is within the inter-observer grading variability levels across the pathologists in our study, which are also consistent with those reported in the literature.

Prostate Biopsy Markers of Inflammation are Associated with Risk of Clinical Progression of Benign Prostatic Hyperplasia: Findings from the MTOPS Study.

PURPOSE: Factors associated with worsening of benign prostatic hyperplasia are not well understood. We measured inflammatory markers from prostate biopsies to study if inflammation is related to clinical progression of benign prostatic hyperplasia. MATERIALS AND METHODS: We measured inflammatory cell markers CD45, CD4, CD8 and CD68 in transition zone biopsies from 859 men in the MTOPS biopsy substudy. Using novel imaging techniques we quantified amounts of moderate/severe inflammation. Benign prostatic hyperplasia clinical progression was defined as a confirmed 4-point or greater increase in the AUA symptom score from baseline, or the occurrence of urinary incontinence or acute urinary retention. Baseline clinical parameters including concomitant medication use were determined. Kaplan-Meier curves and multivariate Cox proportional hazard models were used to determine the risk of progression. RESULTS: Inflammation as measured by CD45, CD4 and CD68 increased the risk of clinical progression of benign prostatic hyperplasia. CD4 showed the highest risk where men in the highest tertile of moderate/severe inflammation were at twice the risk of progression compared to men in the lower 2 tertiles combined (HR 2.03, p=0.001). Inflammation was more strongly associated with progression defined by acute urinary retention or incontinence (HR ranging from 2.39 [CD8, p=0.03] to 3.08 [CD4, p=0.01]) than an AUA symptom score increase (CD4, HR 1.86, p=0.01). Men who reported use of nonsteroidal anti-inflammatory drugs or steroids at baseline tended to be at higher risk for progression. CONCLUSIONS: Although our data show that inflammation increases the risk of progression, our findings suggest that inflammation has a greater role in men who have conditions requiring anti-inflammatory medications.

Reduced expression of GDF-15 is associated with atrophic inflammatory lesions of the prostate.

BACKGROUND: Accumulating evidence suggests that chronic prostatic inflammation may lead to prostate cancer development. Growth differentiation factor-15 (GDF-15) is highly expressed in the prostate and has been associated with inflammation and tumorigenesis. METHODS: To examine the relationship between GDF-15 and prostatic inflammation, GDF-15 expression was measured by immunohistochemical (IHC) staining in human prostatectomy specimens containing inflammation. The relationship between GDF-15 and specific inflammatory cells was determined using non-biased computer image analysis. To provide insight into a potential suppressive role for GDF-15 in inflammation, activation of inflammatory mediator nuclear factor of kappa B (NFκB) was measured in PC3 cells. RESULTS: GDF-15 expression in luminal epithelial cells was decreased with increasing inflammation severity, suggesting an inverse association between GDF-15 and inflammation. Quantification of IHC staining by image analysis for GDF-15 and inflammatory cell markers revealed an inverse correlation between GDF-15 and CD3+, CD4+, CD8+, CD68+, and inos+ leukocytes. GDF-15 suppressed NFκB activity in luciferase reporter assays. Expression of the NFκB target, interleukin 8 (IL-8), was downregulated by GDF-15. CONCLUSIONS: The inverse relationship between GDF-15 and inflammation demonstrates a novel expression pattern for GDF-15 in the human prostate and suppression of NFκB activity may shed light on a potential mechanism for this inverse correlation.

Prostatic atrophy: its spatial proximity to carcinoma and intraepithelial neoplasia based on annotation of digital slides.

Whether atrophy is a precursor to high-grade prostatic intraepithelial neoplasia (HGPIN) and cancer is controversial. A virtual slide set comprising 48 prostatectomy cases was used to investigate associations among the amounts and spacing of these entities. Foci of atrophy without inflammation (A), atrophy with inflammation (AI), cancer (by patterns), and HGPIN were digitally annotated. Atrophy's proximity to cancer and HGPIN was assessed with two measurements: abutment (touching) or nearness (≤2 µm without touching). Area sums per specimen were computed for A, AI, cancer, and HGPIN. Abutment rates of AI and A foci to cancer were 23% versus 21% (p = NS); for nearness, 29% of AI foci were near to cancer versus 12% of A (P = .0001). Abutment or nearness of A and AI to HGPIN were in the 1.4% to 2.4% range. When A, AI, or HGPIN abutted cancer, it was disproportionately to Gleason grade 3 cancer foci even after adjusting for the lesser frequency of higher-grade cancer foci. Area sums of A, AI, or (A + AI) per specimen showed no correlations with those of HGPIN, and mostly negative ones with area sum and with tumor volume of cancer. In conclusion, atrophy with inflammation showed some preferential spatial association to cancer, although area sums of atrophy with or without inflammation correlated negatively with those of cancer. These divergent spatial associations suggest that atrophy and inflammation in biopsy specimens may have clinical relevance. The frequency of inflammatory atrophy (AI) merging with HGPIN was far less than reported previously, weakening the theory that AI gives rise to HGPIN.

Pseudolumen size and perimeter in prostate cancer: correlation with patient outcome.

We demonstrated in 2011 that 61% of men with postoperative PSA failure had some cribriform pattern of prostate cancer, versus 16% of nonfailures (OR = 5.89, P < .0001). That study used digitized radical prostatectomy slides from 153 men, 76 failures (≥0.2 ng/mL) matched to 77 nonfailures. The current study's hypothesis: pseudolumen size and shape variability could stratify outcome within histologic patterns (single separate acini, separate acini with undulating lumens, fused small acini, papillary, cribriform). Pseudolumens were filled digitally on image captures from previously annotated specimens. Among all 5 patterns, pseudolumen spaces averaged smaller in failures than nonfailures. After multivariate analysis controlling for stage, age, margin, cancer amount, prostate volume, and presence of individual cells (grade 5), this retained significance only for the undulating-lumens and papillary patterns. In undulating-lumens pattern, PSA failures had smaller mean pseudolumen space sizes (P = .03) but larger perimeters (P = .04), implying more pseudolumen irregularity. In papillary pattern, the number of pseudolumen spaces was higher in failures (P = .015), space size was smaller (P = .11), perimeters were smaller (P = .04), and perimeter/size ratio was higher (P = .02). In conclusion, digitally measured pseudolumen size and shape may associate with outcome.

Digital quantification of five high-grade prostate cancer patterns, including the cribriform pattern, and their association with adverse outcome.

Proper grading of the cribriform prostate cancer pattern has not previously been supported by outcome-based evidence. Among 153 men who underwent radical prostatectomy, 76 with prostate-specific antigen (PSA) failure (≥0.2 ng/mL [0.2 µg/L]) were matched to 77 without failure. Frequencies of high-grade patterns included fused small acini, 83.7%; papillary, 52.3%; large cribriform, 37.9%; small (≤12 lumens) cribriform, 17.0%; and individual cells, 22.9%. A cribriform pattern was present in 61% (46/76) of failures but 16% (12/77) of nonfailures (P < .0001). Multivariate analysis showed the cribriform pattern had the highest odds ratio for PSA failure, 5.89 (95% confidence interval, 2.53-13.70; P < .0001). The presence of both large and small cribriform patterns was significantly linked to failure. The cumulative odds ratio of failure per added square millimeter of cribriform pattern was 1.173 (P = .008), higher than for any other pattern. All 8 men with a cribriform area sum of 25 mm(2) or more had failure (range, 33-930). Regrading cribriform cancer as Gleason 5 improved the grade association with failure, although half of all cases with individual cells also had a cribriform pattern, precluding a precise determination of the independent importance of the latter. The cribriform pattern has particularly adverse implications for outcome.

Alternatively activated macrophages and collagen remodeling characterize the postpartum involuting mammary gland across species.

Recent pregnancy correlates with decreased survival for breast cancer patients compared with non-pregnancy-associated breast cancer. We hypothesize that postpartum mammary involution induces metastasis through wound-healing programs known to promote cancer. It is unknown whether alternatively activated M2 macrophages, immune cells important in wound-healing and experimental tumorigenesis that also predict poor prognosis for breast cancer patients, are recruited to the normal involuting gland. Macrophage markers CD68, CSF-1R, and F4/80 were examined across the pregnancy and involution cycle in rodent and human mammary tissues. Quantitative immunohistochemistry revealed up to an eightfold increase in macrophage number during involution, which returned to nulliparous levels with full regression. The involution macrophages exhibit an M2 phenotype as determined by high arginase-1 and low inducible nitric oxide synthase staining in rodent tissue, and by mannose receptor expression in human breast tissue. M2 cytokines IL-4 and IL-13 also peaked during involution. Extracellular matrix (ECM) isolated from involuting rat mammary glands was chemotactic for macrophages compared with nulliparous mammary ECM. Fibrillar collagen levels and proteolysis increased dramatically during involution, and denatured collagen I acted as a strong chemoattractant for macrophages in cell culture, suggesting proteolyzed fibrillar collagen as a candidate ECM mediator of macrophage recruitment. M2 macrophages, IL-4, IL-13, fibrillar collagen accumulation, and proteolysis of collagen are all components of tumor promotional microenvironments, and thus may mediate promotion of breast cancers arising in the postpartum setting.

Tamoxifen induces pleiotrophic changes in mammary stroma resulting in extracellular matrix that suppresses transformed phenotypes.

INTRODUCTION: The functional unit of the mammary gland has been defined as the epithelial cell plus its microenvironment, a hypothesis that predicts changes in epithelial cell function will be accompanied by concurrent changes in mammary stroma. To test this hypothesis, the question was addressed of whether mammary stroma is functionally altered by the anti-oestrogen drug tamoxifen. METHODS: Forty female rats at 70 days of age were randomised to two groups of 20 and treated with 1.0 mg/kg tamoxifen or vehicle subcutaneously daily for 30 days, followed by a three-day wash out period. Mammary tissue was harvested and effects of tamoxifen on mammary epithelium and stroma determined. RESULTS: As expected, tamoxifen suppressed mammary alveolar development and mammary epithelial cell proliferation. Primary mammary fibroblasts isolated from tamoxifen-treated rats displayed a three-fold decrease in motility and incorporated less fibronectin in their substratum in comparison to control fibroblasts; attributes indicative of fibroblast quiescence. Immunohistochemistry analysis of CD68, a macrophage lysosomal marker, demonstrated a reduction in macrophage infiltration in mammary glands of tamoxifen-treated rats. Proteomic analyses by mass spectrometry identified several extracellular matrix (ECM) proteins with expression levels with tamoxifen treatment that were validated by Western blot. Mammary tissue from tamoxifen-treated rats had decreased fibronectin and increased collagen 1 levels. Further, ECM proteolysis was reduced in tamoxifen-treated rats as detected by reductions in fibronectin, laminin 1, laminin 5 and collagen 1 cleavage fragments. Consistent with suppression in ECM proteolysis with tamoxifen treatment, matrix metalloproteinase-2 levels and activity were decreased. Biochemically extracted mammary ECM from tamoxifen-treated rats suppressed in vitro macrophage motility, which was rescued by the addition of proteolysed collagen or fibronectin. Mammary ECM from tamoxifen-treated rats also suppressed breast tumour cell motility, invasion and haptotaxis, reduced organoid size in 3-dimensional culture and blocked tumour promotion in an orthotopic xenograft model; effects which could be partially reversed by the addition of exogenous fibronectin. CONCLUSIONS: These data support the hypothesis that mammary stroma responds to tamoxifen treatment in concert with the epithelium and remodels to a microenvironment inhibitory to tumour cell progression. Reduced fibronectin levels and reduced ECM turnover appear to be hallmarks of the quiescent mammary microenvironment. These data may provide insight into attributes of a mammary microenvironment that facilitate tumour dormancy.

Clinical staging of prostate cancer: a computer-simulated study of transperineal prostate biopsy.

OBJECTIVE: To identify the precise location of prostate cancer within the gland and thus possibly permit more aggressive therapy of the lesion, while potentially sparing the noncancerous gland from ablative therapy. MATERIALS AND METHODS: Three-dimensional "solid" computer models were reconstructed for 86 autopsy specimens and 20 stage T1c radical prostatectomy specimens. Transperineal biopsies were simulated for grid sizes of 5-mm (method A) and 10-mm (method B) with an 18 G, 23-mm long biopsy needle. One or two biopsies per grid point were obtained for a total of 12-108 biopsies, depending on the size of the prostate. Clinically threatening cancers were defined as having volumes of > or = 0.5 mL or Gleason sum > or = 7. RESULTS: Method A detected significantly more carcinomas than method B in both the autopsy and prostatectomy specimens (autopsy, 72 vs 51; prostatectomy, 50 vs 32, both P < 0.001). Method A also detected more clinically threatening cancers found at autopsy (38/40 vs 31/40, P = 0.008). Among autopsy patients with negative sextant biopsies whose disease was localized to one side, method A detected 72% and method B detected 29-43% (P < 0.001). CONCLUSIONS: The results of this computer simulation show that 5- and 10-mm grid biopsies detect three-quarters and a third, respectively, at autopsy, of patients with the disease localized to one side of the prostate, which may be useful when planning highly selective ablative treatments in the future.