Crowdsourcing image annotation for nucleus detection and segmentation in computational pathology: evaluating experts, automated methods, and the crowd.

The development of tools in computational pathology to assist physicians and biomedical scientists in the diagnosis of disease requires access to high-quality annotated images for algorithm learning and evaluation. Generating high-quality expert-derived annotations is time-consuming and expensive. We explore the use of crowdsourcing for rapidly obtaining annotations for two core tasks in com- putational pathology: nucleus detection and nucleus segmentation. We designed and implemented crowdsourcing experiments using the CrowdFlower platform, which provides access to a large set of labor channel partners that accesses and manages millions of contributors worldwide. We obtained annotations from four types of annotators and compared concordance across these groups. We obtained: crowdsourced annotations for nucleus detection and segmentation on a total of 810 images; annotations using automated methods on 810 images; annotations from research fellows for detection and segmentation on 477 and 455 images, respectively; and expert pathologist-derived annotations for detection and segmentation on 80 and 63 images, respectively. For the crowdsourced annotations, we evaluated performance across a range of contributor skill levels (1, 2, or 3). The crowdsourced annotations (4,860 images in total) were completed in only a fraction of the time and cost required for obtaining annotations using traditional methods. For the nucleus detection task, the research fellow-derived annotations showed the strongest concordance with the expert pathologist- derived annotations (F-M =93.68%), followed by the crowd-sourced contributor levels 1,2, and 3 and the automated method, which showed relatively similar performance (F-M = 87.84%, 88.49%, 87.26%, and 86.99%, respectively). For the nucleus segmentation task, the crowdsourced contributor level 3-derived annotations, research fellow-derived annotations, and automated method showed the strongest concordance with the expert pathologist-derived annotations (F-M = 66.41%, 65.93%, and 65.36%, respectively), followed by the contributor levels 2 and 1 (60.89% and 60.87%, respectively). When the research fellows were used as a gold-standard for the segmentation task, all three con- tributor levels of the crowdsourced annotations significantly outperformed the automated method (F-M = 62.21%, 62.47%, and 65.15% vs. 51.92%). Aggregating multiple annotations from the crowd to obtain a consensus annotation resulted in the strongest performance for the crowd-sourced segmentation. For both detection and segmentation, crowd-sourced performance is strongest with small images (400 × 400 pixels) and degrades significantly with the use of larger images (600 × 600 and 800 × 800 pixels). We conclude that crowdsourcing to non-experts can be used for large-scale labeling microtasks in computational pathology and offers a new approach for the rapid generation of labeled images for algorithm development and evaluation.

Event excess in the MiniBooNE search for ¯νµâ†’¯νe oscillations.

The MiniBooNE experiment at Fermilab reports results from a search for ¯ν_{µ}→¯ν_{e} oscillations, using a data sample corresponding to 5.66×10²° protons on target. An excess of 20.9±14.0 events is observed in the energy range 475

Search for electron antineutrino appearance at the deltam(2) approximately 1 eV(2) Scale.

The MiniBooNE Collaboration reports initial results from a search for nu(mu)-->nu(e) oscillations. A signal-blind analysis was performed using a data sample corresponding to 3.39x10(20) protons on target. The data are consistent with background prediction across the full range of neutrino energy reconstructed assuming quasielastic scattering, 200

Measurement of the ratio of the numu charged-current single-pion production to quasielastic scattering with a 0.8 GeV neutrino beam on mineral oil.

Using high statistics samples of charged-current numu interactions, the MiniBooNE [corrected] Collaboration reports a measurement of the single-charged-pion production to quasielastic cross section ratio on mineral oil (CH2), both with and without corrections for hadron reinteractions in the target nucleus. The result is provided as a function of neutrino energy in the range 0.4 GeV

Unexplained excess of electronlike events from a 1-GeV neutrino beam.

The MiniBooNE Collaboration observes unexplained electronlike events in the reconstructed neutrino energy range from 200 to 475 MeV. With 6.46x10;{20} protons on target, 544 electronlike events are observed in this energy range, compared to an expectation of 415.2+/-43.4 events, corresponding to an excess of 128.8+/-20.4+/-38.3 events. The shape of the excess in several kinematic variables is consistent with being due to either nu_{e} and nu[over ]_{e} charged-current scattering or nu_{mu} neutral-current scattering with a photon in the final state. No significant excess of events is observed in the reconstructed neutrino energy range from 475 to 1250 MeV, where 408 events are observed compared to an expectation of 385.9+/-35.7 events.

Building epithelial tissues from skin stem cells.

The skin epidermis and its appendages provide a protective barrier that guards against loss of fluids, physical trauma, and invasion by harmful microbes. To perform these functions while confronting the harsh environs of the outside world, our body surface undergoes constant rejuvenation through homeostasis. In addition, it must be primed to repair wounds in response to injury. The adult skin maintains epidermal homeostasis, hair regeneration, and wound repair through the use of its stem cells. What are the properties of skin stem cells, when do they become established during embryogenesis, and how are they able to build tissues with such remarkably distinct architectures? How do stem cells maintain tissue homeostasis and repair wounds and how do they regulate the delicate balance between proliferation and differentiation? What is the relationship between skin cancer and mutations that perturbs the regulation of stem cells? In the past 5 years, the field of skin stem cells has bloomed as we and others have been able to purify and dissect the molecular properties of these tiny reservoirs of goliath potential. We report here progress on these fronts, with emphasis on our laboratory's contributions to the fascinating world of skin stem cells.

Telomerase, cervical cancer, and human papillomavirus.

Review of the available data indicates that telomerase is activated in the majority of cervical squamous cell carcinomas as it is in most malignant neoplasms. Telomerase activity can also be detected in some preneoplastic cervical lesions, but the significance of this in unclear, because nonneoplastic, proliferating epithelial cells also can have telomerase activity. The bias introduced by cytologic sampling methods can complicate the interpretation of results. Quantitative telomerase assays may be useful in distinguishing nonmalignant, physiologic activation of telomerase from malignant activation. Studies evaluating telomerase component (hTR or hTERT) expression by evaluation of RNA, mRNA, or antigen have yielded conflicting results, but the observation that many nonmalignant, nontelomerase active cells have detectable hTR and hTERT suggests that many cells express telomerase RNA and catalytic components, but do not have active telomerase. The implication is that a regulatory overlay must exist that controls telomerase activation. Activation of the enzyme in carcinogenesis could conceivably be a physiologic activation that normally accompanies cellular proliferation, a direct appropriation of telomerase activity by the neoplastic process, or both. The presence of inactive telomerase in many cells also raises the possibility of a noncatalytic function for the telomerase complex. An understanding of telomerase interaction with HPV infection in the pathogenesis of cervical neoplasia must await a further elaboration of telomerase regulation. Likewise, application of telomerase detection in cervical cancer screening programs must await a better integration of telomerase regulation in normal and specifically in HPV-infected squamous epithelial cells.

Estimates of Helicobacter pylori densities in the gastric mucus layer by PCR, histologic examination, and CLOtest.

Helicobacter pylori inhabits the gastric mucus layer of infected persons. A number of investigators have reported the feasibility of detecting H pylori in gastric mucus with polymerase chain reaction (PCR)-based methods. We have established the sensitivity of a simple PCR assay for detecting H pylori in gastric mucus samples and estimate that the density of H pylori organisms in the gastric mucus of untreated patients is approximately 107 to 108 organisms per milliliter. We have similarly estimated the analytic sensitivities of histologic examination and the CLOtest (TRI-MED Specialties, Overland Park, Kan) for detecting H pylori and calculate similar values for the numbers of organisms in the gastric mucus layer. Our data indicate that gastric mucus is a suitable specimen for the detection of H pylori in infected patients, and that PCR-based assays of gastric mucus are significantly more sensitive than histologic testing or the CLOtest for demonstration of H pylori infection.

Deletions of chromosome 3p are frequent and early events in the pathogenesis of uterine cervical carcinoma.

To study the molecular abnormalities involved in the multistage development of cervical carcinoma (CC), we investigated the presence of oncogenic human papillomavirus (HPV) sequences, loss of heterozygosity (LOH), and microsatellite alterations at several genes/loci at 3p (3p14.2 at the FHIT gene, 3p14.3-21.1, 3p21, and 3p22-24.2), 9p21, RB and P53, and P53 gene point mutations in precisely microdissected archival tissues from 20 CCs and their accompanying precursor lesions (cervical intraepithelial neoplasia, CIN; n = 40) and normal epithelia (n = 20). In all HPV-positive cases (90% of CCs), HPV sequences were detected as the earliest appearing molecular change or simultaneously with other changes. LOH at any 3p region was found in 70% of CCs, and 3p14.2 (FHIT gene/FRA3B fragile site) (56%) and 3p21 (57%) were the most frequent 3p sites of loss. LOH at some 3p region was in the CIN I stage, and the 3p deletions in precursor CIN lesions were smaller than the 3p losses found in the associated invasive CC. LOH at the other regions studied and P53 gene mutations were less frequent and later events. Microsatellite alterations were detected in 35% of CCs, and identical abnormalities were detected in the associated precursor lesions. Although infection with oncogenic HPV strains is the earliest and most frequent molecular event, progressive deletions at one or more 3p regions (particularly at 3p14.2, and 3p21) are also frequent events occurring early in the pathogenesis of CC.

Acute primary human immunodeficiency virus type 1 infection in a patient with concomitant cytomegalovirus encephalitis.

We report what we believe is the first case of primary human immunodeficiency virus type 1 (HIV-1) infection and simultaneous cytomegalovirus (CMV) encephalitis, which was confirmed by detection of CMV DNA in the patient's cerebrospinal fluid with use of the polymerase chain reaction. This coinfection had an unusual course, and the patient's clinical status deteriorated despite administration of combination antiretroviral therapy. The patient responded clinically only after therapy for CMV infection was added to his combination antiretroviral regimen. An atypical course and duration of symptomatic primary HIV-1 infection should suggest a possible coincident infection with other opportunistic agents that are normally expected to cause disease later in the course of HIV-1 infection. Current recommendations from the Centers for Disease Control and Prevention list CMV encephalitis as an AIDS-defining event.

p53 immunostaining positivity is associated with reduced survival and is imperfectly correlated with gene mutations in resected non-small cell lung cancer. A preliminary report of LCSG 871.

We investigated the correlation of p53 abnormalities with survival in 85 patients with non-small cell lung cancer (NSCLC) who had undergone resection with curative intent as part of Lung Cancer Study Group (LCSG) 871. Our previous studies showed that only a subset of p53 mutations in lung cancers result in overexpression. In addition, protein overexpression has been described in the absence of mutation. Therefore, we determined both p53 protein overexpression (by immunostaining) and p53 and ras gene mutations (by single-strand conformation polymorphism and DNA sequencing) in this set of resected tumor specimens. Clinical follow-up data were available for 75 cases. Of the studied patients, 64% showed p53 overexpression and 51% had mutant p53 sequences; however, the concordance rate was only 67%. There was a negative survival correlation with positive p53 immunostaining (p = 0.05), but not with the presence of gene mutations (p = 0.62) in this group of patients. Overexpression of p53 protein determined by immunostaining may contribute to adverse outcome due to the ability of p53 to act as a dominant oncogene, or alternatively, overexpression may reflect ongoing DNA damage in the tumor as a marker for a more aggressive behavior. When adjusted for stage, age, and gender by multivariate analysis, however, there was no independent impact of p53 overexpression on survival.

Primary lymphoma of the liver. Report of a case with diagnosis by fine needle aspiration.

In a 69-year-old man with hepatomegaly, a diagnosis of primary non-Hodgkin's lymphoma (NHL) of the liver was made by fine needle aspiration (FNA). At the time of presentation there was no evidence of involvement of the lymph nodes, bone marrow or any other organ. Although hepatic involvement is common in advanced stages of Hodgkin's disease and NHL, primary lymphoma of the liver is rare. The purpose of this paper is to report a rare occurrence of primary lymphoma of the liver and to demonstrate the possibility of making this diagnosis by FNA.

Interaction of bovine erythrocyte N-glycolylneuraminic acid-containing gangliosides and glycoproteins with a human Hanganutziu-Deicher serum.

A human Hanganutziu-Deicher (H-D) serum reacted with N-glycolylneuraminic acid (NeuGc)-containing gangliosides and glycoproteins isolated from bovine erythrocyte membranes. Three populations of H-D antibodies were identified in the human H-D serum. One population very likely recognized the NeuGc-Gal sequence; a second population appears to recognize additional sugars in the oligosaccharide sequence, e.g. NeuGc-Gal-GlcNAc; while a third population may also recognize polypeptide determinants in addition to the NeuGc-Gal-GlcNAc sequence. The H-D serum distinguished two high mol. wt glycoproteins (HMGP I and II) present in crude extracts of bovine erythrocyte membranes. These glycoproteins were separated by repetitive fractionation on Sephacryl S-1000 in the presence of urea and their composition determined.

Mycoplasma restriction: identification of a new type of restriction specificity for DNA containing 5-methylcytosine.

Mycoplasma bacteriophage L51 single-stranded DNA and L2 double-stranded DNA are host cell modified and restricted when they transfect Acholeplasma laidlawii JA1 and K2 cells. The L51 genome has a single restriction endonuclease MboI site (recognition sequence GATC), which contains 5-methylcytosine when the DNA is isolated from L51 phage grown in K2 cells but is unmethylated when the DNA is from phage grown in JA1 cells. This GATC sequence is nonessential, since an L51 mutant in which the MboI site was deleted was still viable. DNA from this deletion mutant phage was not restricted during transfection of either strain K2 or JA1. Therefore, strain K2 restricts DNA containing the sequence GATC, and strain JA1 restricts DNA containing the sequence GAT 5-methylcytosine. We conclude that K2 cells have a restriction system specific for DNA containing the sequence GATC and protect their DNA by methylating cytosine in this sequence. In contrast, JA1 cells (which contain no methylated DNA bases) have a newly discovered type of restriction-modification system. From results of studies of the restriction of specifically methylated DNAs, we conclude that JA1 cells restrict DNA containing 5-methylcytosine, regardless of the nucleotide sequence containing 5-methylcytosine. This is the first report of a DNA restriction activity specific for a single (methylated) base. Modification in this system is the absence of cytosine methylating activity. A restriction-deficient variant of strain JA1, which retains the JA1 modification phenotype, was isolated, indicating that JA1 cells have a gene product with restriction specificity for DNA containing 5-methylcytosine.

Monoclonal antibodies to human renal cell carcinoma: recognition of shared and restricted tissue antigens.

Monoclonal antibodies (MABs) reactive with human renal cell carcinoma (RCC) were generated following immunization of mice with either RCC homogenates, RCC cell lines, or fetal kidney homogenates. The characteristics of four highly reactive immunoglobulin G1 MABs, designated UMVA-RCC-A6H, UMVA-RCC-A36, UMVA-RCC-C5H and UMVA-RCC-D5D are presented. The screening process consisted of a cell binding enzyme-linked immunosorbent assay and immunohistological examination of tumor, normal, and fetal tissue sections. The MABs illustrated various degrees of antigen restriction: A6H identified an antigen common to RCC, some lung and colon carcinomas, the proximal renal tubules but no other normal tissues; A36 reacted with most human tumors, the renal tubules, and many other normal tissues; C5H reacted with nearly every human cancer but of the normal tissues, only the renal glomerulus shared this antigen; D5D was very restrictive, reacting with many although not all RCC and no other cancers or normal tissues with the exception of an occasional reactivity with a Bowman's capsule. Metastatic RCC and RCC xenografts expressed these antigens. None of the MABs participated in complement-mediated cytotoxicity. In immunoprecipitation studies with L-[methyl-3H]methionine and [3H]glucosamine-HCl metabolically labeled RCC cells, C5H was shown to be associated with an antigen of Mr 115,000.

Identification of an enveloped phage, mycoplasma virus L172, that contains a 14-kilobase single-stranded DNA genome.

We have found that mycoplasma virus L172 is an enveloped globular virion containing circular, single-stranded DNA of 14.0 kilobases. L172 has been reported by other workers to have a double-stranded DNA genome of 13 to 17 kilobase pairs and has been classified as a plasmavirus, a group for which mycoplasma virus L2 is the type member. Mycoplasma viruses L172 and L2 differ in genome size and structure, DNA base composition, and protein composition, and they have no detectable DNA homology. As the only reported enveloped virion containing single-stranded DNA, L172 represents a new group of viruses.

A highly restricted antigen for renal cell carcinoma defined by a monoclonal antibody.

Monoclonal antibodies against renal cell carcinoma (RCC) antigens were generated by immunizing Balb/c mice with the human RCC cell line 7860. After cloning many RCC reactive monoclonal antibodies, one antibody (D5D), was of special interest. Specificity testing against 33 human tumor cell lines revealed D5D to have no specific reactivity with 17 specimens of normal adult or fetal kidney. Reactivity with 67 specimens of nonrenal tissue demonstrated no reactivity. The reactivity of D5D with 15 different RCC tissues has been variable, being present in 60% of cases. Preliminary data, however, suggest that the antigen may be reexpressed for some of the remaining 40% during in vitro propagation.

Transposon-mediated conjugational transmission of nonconjugative plasmids.

When coresident with conjugative plasmid pNC21, the nonconjugative deletion F-prime pJC59, which retains the F transfer origin oriT, was transmitted to transconjugants at a frequency comparable to that of pNC21. In addition, pJC59 was transmitted as an independent plasmid, physically separate from pNC21, an example of plasmid donation. In contrast, two plasmids that are derived from F and deleted for the oriT site, pJC61 and pML31, were transmitted at frequencies 10(4) lower than that of pNC21. This low-frequency transmission was associated with the appearance of a new plasmid in the transconjugants. In the case of pML31, we determined that this new plasmid was a recombinant composed of pNC21 and pML31, the latter flanked by two copies of transposable element Tn3. We believe that this recombinant plasmid was formed as an intermediate in the transposition of Tn3 from pNC21 to pML31 and was the vehicle for conjugational transmission of pML31 genes by a process known as plasmid conduction.