Deep Understanding of Breast Density Classification.

We have developed a deep learning architecture, DualViewNet, for mammogram density classification as well as a novel metric for quantifying network preference of mediolateral oblique (MLO) versus craniocaudal (CC) views in density classification. Also, we have provided thorough analysis and visualization to better understand the behavior of deep neural networks in density classification. Our proposed architecture, DualViewNet, simultaneously examines and classifies both MLO and CC views corresponding to the same breast, and shows best performance with a macro average AUC of 0.8970 and macro average 95% confidence interval of 0.8239-0.9450 obtained via bootstrapping 1000 test sets. By leveraging DualViewNet we provide a novel algorithm and quantitative comparison of MLO versus CC views for classification and find that MLO provides stronger influence in 1,187 out of 1,323 breasts.

MAPGI: Accurate identification of anatomical landmarks and diseased tissue in gastrointestinal tract using deep learning.

Automatic detection of anatomical landmarks and diseases in medical images is a challenging task which could greatly aid medical diagnosis and reduce the cost and time of investigational procedures. Also, two particular challenges of digital image processing in medical applications are the sparsity of annotated medical images and the lack of uniformity across images and image classes. This paper presents methodologies for maximizing classification accuracy on a small medical image dataset, the Kvasir dataset, by performing robust image preprocessing and applying state-of-the-art deep learning. Images are classified as being or involving an anatomical landmark (pylorus, z-line, cecum), a diseased state (esophagitis, ulcerative colitis, polyps), or a medical procedure (dyed lifted polyps, dyed resection margins). A framework for modular and automatic preprocessing of gastrointestinal tract images (MAPGI) is proposed, which applies edge removal, contrast enhancement, filtering, color mapping and scaling to each image in the dataset. Gamma correction values are automatically calculated for individual images such that the mean pixel value for each image is normalized to 90 ±â€¯1 in a 0-255 pixel value range. Three state-of-the-art neural networks architectures, Inception-ResNet-v2, Inception-v4, and NASNet, are trained on the Kvasir dataset, and their classification performance is juxtaposed on validation data. In each case, 85% of the images from the Kvasir dataset are used for training, while the other 15% are reserved for validation. The resulting accuracies achieved using Inception-v4, Inception-ResNet-v2, and NASNet were 0.9845, 0.9848, and 0.9735, respectively. In addition, Inception-v4 achieved an average of 0.938 precision, 0.939 recall, 0.991 specificity, 0.938 F1 score, and 0.929 Matthews correlation coefficient (MCC). Bootstrapping provided NASNet, the worst performing model, a lower bound of 0.9723 accuracy on the 95% confidence interval.

RAMS: Remote and automatic mammogram screening.

About one in eight women in the U.S. will develop invasive breast cancer at some point in life. Breast cancer is the most common cancer found in women and if it is identified at an early stage by the use of mammograms, x-ray images of the breast, then the chances of successful treatment can be high. Typically, mammograms are screened by radiologists who determine whether a biopsy is necessary to ascertain the presence of cancer. Although historical screening methods have been effective, recent advances in computer vision and web technologies may be able to improve the accuracy, speed, cost, and accessibility of mammogram screenings. We propose a total screening solution comprised of three main components: a web service for uploading images and reviewing results, a machine learning algorithm for accepting or rejecting images as valid mammograms, and an artificial neural network for locating potential malignancies. Once an image is uploaded to our web service, an image acceptor determines whether or not the image is a mammogram. The image acceptor is primarily a one-class SVM built on features derived with a variational autoencoder. If an image is accepted as a mammogram, the malignancy identifier, a ResNet-101 Faster R-CNN, will locate tumors within the mammogram. On test data, the image acceptor had only 2 misclassifications out of 410 mammograms and 2 misclassifications out of 1,640 non-mammograms while the malignancy identifier achieved 0.951 AUROC when tested on BI-RADS 1, 5, and 6 images from the INbreast dataset.

Biodiversity of the Surface Microbial Consortia from Limburger, Reblochon, Livarot, Tilsit, and Gubbeen Cheeses.

Comprehensive collaborative studies from our laboratories reveal the extensive biodiversity of the microflora of the surfaces of smear-ripened cheeses. Two thousand five hundred ninety-seven strains of bacteria and 2,446 strains of yeasts from the surface of the smear-ripened cheeses Limburger, Reblochon, Livarot, Tilsit, and Gubbeen, isolated at three or four times during ripening, were identified; 55 species of bacteria and 30 species of yeast were found. The microfloras of the five cheeses showed many similarities but also many differences and interbatch variation. Very few of the commercial smear microorganisms, deliberately inoculated onto the cheese surface, were reisolated and then mainly from the initial stages of ripening, implying that smear cheese production units must have an adventitious "house" flora. Limburger cheese had the simplest microflora, containing two yeasts, Debaryomyces hansenii and Geotrichum candidum, and two bacteria, Arthrobacter arilaitensis and Brevibacterium aurantiacum. The microflora of Livarot was the most complicated, comprising 10 yeasts and 38 bacteria, including many gram-negative organisms. Reblochon also had a very diverse microflora containing 8 yeasts and 13 bacteria (excluding gram-negative organisms which were not identified), while Gubbeen had 7 yeasts and 18 bacteria and Tilsit had 5 yeasts and 9 bacteria. D. hansenii was by far the dominant yeast, followed in order by G. candidum, Candida catenulata, and Kluyveromyces lactis. B. aurantiacum was the dominant bacterium and was found in every batch of the 5 cheeses. The next most common bacteria, in order, were Staphylococcus saprophyticus, A. arilaitensis, Corynebacterium casei, Corynebacterium variabile, and Microbacterium gubbeenense. S. saprophyticus was mainly found in Gubbeen, and A. arilaitensis was found in all cheeses but not in every batch. C. casei was found in most batches of Reblochon, Livarot, Tilsit, and Gubbeen. C. variabile was found in all batches of Gubbeen and Reblochon but in only one batch of Tilsit and in no batch of Limburger or Livarot. Other bacteria were isolated in low numbers from each of the cheeses, suggesting that each of the 5 cheeses has a unique microflora. In Gubbeen cheese, several different strains of the dominant bacteria were present, as determined by pulsed-field gel electrophoresis, and many of the less common bacteria were present as single clones. The culture-independent method, denaturing gradient gel electrophoresis, resulted in identification of several bacteria which were not found by the culture-dependent (isolation and rep-PCR identification) method. It was thus a useful complementary technique to identify other bacteria in the cheeses. The gross composition, the rate of increase in pH, and the indices of proteolysis were different in most of the cheeses.

Commercial ripening starter microorganisms inoculated into cheese milk do not successfully establish themselves in the resident microbial ripening consortia of a South german red smear cheese.

Production of smear-ripened cheese critically depends on the surface growth of multispecies microbial consortia comprising bacteria and yeasts. These microorganisms often originate from the cheese-making facility and, over many years, have developed into rather stable, dairy-specific associations. While commercial smear starters are frequently used, it is unclear to what degree these are able to establish successfully within the resident microbial consortia. Thus, the fate of the smear starters of a German Limburger cheese subjected to the "old-young" smearing technique was investigated during ripening. The cheese milk was supplemented with a commercial smear starter culture containing Debaryomyces hansenii, Galactomyces geotrichum, Arthrobacter arilaitensis, and Brevibacterium aurantiacum. Additionally, the cheese surface was inoculated with an extremely stable in-house microbial consortium. A total of 1,114 yeast and 1,201 bacterial isolates were identified and differentiated by Fourier transform infrared spectroscopy. Furthermore, mitochondrial DNA restriction fragment length polymorphism, random amplified polymorphic DNA, repetitive PCR, and pulsed field gel electrophoresis analyses were used to type selected isolates below the species level. The D. hansenii starter strain was primarily found early in the ripening process. The G. geotrichum starter strain in particular established itself after relocation to a new ripening room. Otherwise, it occurred at low frequencies. The bacterial smear starters could not be reisolated from the cheese surface at all. It is concluded that none of the smear starter strains were able to compete significantly and in a stable fashion against the resident microbial consortia, a result which might have been linked to the method of application. This finding raises the issue of whether addition of starter microorganisms during production of this type of cheese is actually necessary.

Growth characteristics of Brevibacterium, Corynebacterium, Microbacterium, and Staphylococcus spp. isolated from surface-ripened cheese.

The growth characteristics of five bacteria, Brevibacterium aurantiacum 1-16-58, Corynebacterium casei DPC 5298(T), Corynebacterium variabile DPC 5310, Microbacterium gubbeenense DPC 5286(T), and Staphylococcus saprophyticus 4E61, all of which were isolated from the surface of smear cheese, were studied in complex and chemically defined media. All of the coryneforms, except M. gubbeenense, grew in 12% salt, while B. aurantiacum and S. saprophyticus grew in 15% salt. All five bacteria assimilated lactate in a semisynthetic medium, and none of the coryneform bacteria assimilated lactose. Glucose assimilation was poor, except by S. saprophyticus and C. casei. Five to seven amino acids were assimilated by the coryneforms and 12 by S. saprophyticus. Glutamate, phenylalanine, and proline were utilized by all five bacteria, whereas utilization of serine, threonine, aspartate, histidine, alanine, arginine, leucine, isoleucine, and glycine depended on the organism. Growth of C. casei restarted after addition of glutamate, proline, serine, and lactate at the end of the exponential phase, indicating that these amino acids and lactate can be used as energy sources. Pantothenic acid was essential for the growth of C. casei and M. gubbeenense. Omission of biotin reduced the growth of B. aurantiacum, C. casei, and M. gubbeenense. All of the bacteria contained lactate dehydrogenase activity (with both pyruvate and lactate as substrates) and glutamate pyruvate transaminase activity but not urease activity.

Agrococcus casei sp. nov., isolated from the surfaces of smear-ripened cheeses.

Seven Gram-positive, coryneform bacteria with virtually identical whole-organism protein patterns were isolated from the surface of smear-ripened cheeses. Representatives of these strains were the subject of a polyphasic study designed to establish their taxonomic status. The organisms formed a distinct branch in the Microbacteriaceae 16S rRNA gene tree and were most closely related to members of the genus Agrococcus, sharing sequence similarities of 95.4-98.7 %. The chemotaxonomic profiles of the strains were consistent with their classification in the genus Agrococcus. The combined genotypic and phenotypic data show that the isolates should be classified in the genus Agrococcus as representatives of a novel species. The name Agrococcus casei sp. nov. is proposed for this taxon. Isolate R-17892t2(T) (=DSM 18061(T)=LMG 22410(T)) is the type strain of Agrococcus casei sp. nov.

Enterococcus and Lactobacillus contamination of raw milk in a farm dairy environment.

Enterococci and lactobacilli are ubiquitously found in the intestinal microflora of humans and animals. The aim of the present study was to determine the importance of bovine faeces as a source of these organisms in raw milk. One hundred and fifty six putative enterococci and 362 lactobacilli were isolated from bovine faeces (n=26), cows' teats, raw milk, the milking machine and the milking environment on one farm. The clonal relationships of each group were investigated using Pulsed-Field Gel Electrophoresis and representatives of the different clusters were identified by repetitive DNA element (rep)-PCR fingerprinting, protein profiling, phenylalanyl-tRNA synthase (pheS) sequence analysis or 16S rDNA gene sequencing. Lactobacilli were present at approximately 3 orders of magnitude greater than enterococci in the bovine faeces. The majority of the bovine faecal enterococcal isolates were identified as Aerococcus viridans. Seven teat isolates belonged to a potential novel Aerococcus sp. and one bovine faecal isolate to a potential second novel Aerococcus sp. The lactobacilli present in the bovine faeces were predominantly Lactobacillus mucosae and Lactobacillus brevis, with small numbers of Lactobacillus plantarum. Only one Enterococcus (a strain of E. casseliflavus) out of 76 and one Lactobacillus (a strain of L. parabuchneri/kefir) out of 247 of the bovine faecal isolates was found in the milk. The major source of these bacteria in the milk was the milking equipment.

Growth and colour development of some surface ripening bacteria with Debaryomyces hansenii on aseptic cheese curd.

The growth of five bacteria isolated from red-smear cheeses, Brevibacterium aurantiacum, Corynebacterium casei, Corynebacterium variabile, Microbacterium gubbeenense and Staphylococcus saprophyticus in mixed cultures with Debaryomyces hansenii on aseptic model cheese curd at 10 and 14 degrees C was investigated. At both temperatures, C. casei and Micro. gubbeenense had a longer lag phase than C. variabile, Brevi. aurantiacum and Staph. saprophyticus. In all cultures, lactose was utilised first and was consumed more rapidly at 14 degrees C than at 10 degrees C, i.e., 6 d at 14 degrees C and 10 d at 10 degrees C. This utilisation coincided with the exponential growth of Deb. hansenii on the cheese surface. Lactate was also used as a carbon source and was totally consumed after 21 d at 14 degrees C and approximately 90% was consumed after 21 d at 10 degrees C regardless of the ripening culture. Small differences (<0.5 pH unit) in the surface-pH during ripening were noticeable between ripening cultures. Differences in the colour development of the mixed cultures with the yeast control were only noticeable after 15 d for Brevi. aurantiacum and after 21 d for the other bacteria. Regardless of the organisms tested, colour development and colour intensity were also greater at 14 degrees C than at 10 degrees C. This study has provided useful information on the growth and contribution to colour development of these bacteria on cheese.

Kluyveromyces lactis and Saccharomyces cerevisiae, two potent deacidifying and volatile-sulphur-aroma-producing microorganisms of the cheese ecosystem.

Cheese flavour is the result of complex biochemical transformations attributed to bacteria and yeasts grown on the curd of smear-ripened cheeses. Volatile sulphur compounds (VSCs) are responsible for the characteristic aromatic notes of several cheeses. In the present study, we have assessed the ability of Kluyveromyces lactis, Kluyveromyces marxianus and Saccharomyces cerevisiae strains, which are frequently isolated from smear-ripened cheeses, to grow and deacidify a cheese medium and generate VSCs resulting from L-methionine degradation. The Kluyveromyces strains produced a wider variety and higher amounts of VSCs than the S. cerevisiae ones. We have shown that the pathway is likely to be proceeding differently in these two yeast genera. The VSCs are mainly generated through the degradation of 4-methylthio-oxobutyric acid in the Kluyveromyces strains, in contrast to the S. cerevisiae ones which have higher L-methionine demethiolating activity, resulting in a direct conversion of L-methionine to methanethiol. The deacidification activity which is of major importance in the early stages of cheese-ripening was also compared in S. cerevisiae and Kluyveromyces strains.

L-methionine degradation pathway in Kluyveromyces lactis: identification and functional analysis of the genes encoding L-methionine aminotransferase.

Kluyveromyces lactis is one of the cheese-ripening yeasts and is believed to contribute to the formation of volatile sulfur compounds (VSCs) through degradation of L-methionine. L-methionine aminotransferase is potentially involved in the pathway that results in the production of methanethiol, a common precursor of VSCs. Even though this pathway has been studied previously, the genes involved have never been studied. In this study, on the basis of sequence homology, all the putative aminotransferase-encoding genes from K. lactis were cloned in an overproducing vector, pCXJ10, and their effects on the production of VSCs were analyzed. Two genes, KlARO8.1 and KlARO8.2, were found to be responsible for L-methionine aminotransferase activity. Transformants carrying these genes cloned in the pCXJ10 vector produced threefold-larger amounts of VSCs than the transformant containing the plasmid without any insert or other related putative aminotransferases produced.

Surface microflora of four smear-ripened cheeses.

The microbial composition of smear-ripened cheeses is not very clear. A total of 194 bacterial isolates and 187 yeast isolates from the surfaces of four Irish farmhouse smear-ripened cheeses were identified at the midpoint of ripening using pulsed-field gel electrophoresis (PFGE), repetitive sequence-based PCR, and 16S rRNA gene sequencing for identifying and typing the bacteria and Fourier transform infrared spectroscopy and mitochondrial DNA restriction fragment length polymorphism (mtDNA RFLP) analysis for identifying and typing the yeast. The yeast microflora was very uniform, and Debaryomyces hansenii was the dominant species in the four cheeses. Yarrowia lipolytica was also isolated in low numbers from one cheese. The bacteria were highly diverse, and 14 different species, Corynebacterium casei, Corynebacterium variabile, Arthrobacter arilaitensis, Arthrobacter sp., Microbacterium gubbeenense, Agrococcus sp. nov., Brevibacterium linens, Staphylococcus epidermidis, Staphylococcus equorum, Staphylococcus saprophyticus, Micrococcus luteus, Halomonas venusta, Vibrio sp., and Bacillus sp., were identified on the four cheeses. Each cheese had a more or less unique microflora with four to nine species on its surface. However, two bacteria, C. casei and A. arilaitensis, were found on each cheese. Diversity at the strain level was also observed, based on the different PFGE patterns and mtDNA RFLP profiles of the dominant bacterial and yeast species. None of the ripening cultures deliberately inoculated onto the surface were reisolated from the cheeses. This study confirms the importance of the adventitious, resident microflora in the ripening of smear cheeses.

Corynebacterium mooreparkense, a later heterotypic synonym of Corynebacterium variabile.

Strains of a Gram-positive bacterium were isolated from the Irish smear-ripened cheese Gubbeen, and assigned to a new species, Corynebacterium mooreparkense, in 2001. During a further study on the same cheese, no additional isolates from this species could be found. Instead, multiple isolates of its nearest phylogenetic neighbour, Corynebacterium variabile, were found. A first screening with rep-PCR and SDS-PAGE pointed to a similarity between C. mooreparkense and C. variabile. Following this peculiar result, attempts were made to collect all type strains deposited at different culture collections and all strains described by Brennan et al. [Int J Syst Evol Microbiol (2001) 51, 843-852]. Subsequently, 16S rRNA gene sequencing and DNA-DNA hybridizations were performed. All C. mooreparkense strains had a 16S rRNA gene sequence similarity of at least 99.5 % with C. variabile and the DNA-DNA relatedness was 95 %. On the basis of these results, it is concluded that C. mooreparkense is a later heterotypic synonym of C. variabile.

Glucose prevents citrate metabolism by enterococci.

Enterococcus faecalis FAIR E-239, growing on glucose plus citrate, metabolized citrate at pH 6.5 or 7.5, but only when glucose had been exhausted; it did not metabolize citrate at pH 5.5 or 8.5. When grown on citrate only, the strain metabolized citrate at all pH values, and two growth rates were apparent. Citrate was mainly metabolized during the second, much slower growth rate. Glucose also inhibited citrate metabolism by E. faecalis FAIR E-237 and FAIR E-259 and Enterococcus faecium FAIR E-338 and FAIR E-371. Glucose-grown resting cells were unable to metabolize citrate. Citrate-grown resting cells had a pH optimum of 4.7 for citrate metabolism but also metabolized significant amounts of citrate at pH 4.2 and 6.5. Resting stationary phase cells used citrate more rapidly than resting log phase cells. Citrate metabolism was faster at citrate levels <10 mM than above 10 mM. These results suggest that some form of catabolite repression is occurring.

Catabolite repression in Enterococcus faecalis.

Metabolism of citrate, pyruvate and sugars by Enterococcus faecalis E-239 and JH2-2 and an isogenic, catabolite derepressed mutant of JH2-2, strain CL4, was investigated. The growth rates of E. faecalis E-239 on citrate and pyruvate were 0.58 and 0.63 h(-1), respectively, indicating that both acids were used as energy sources. Fructose and glucose prevented the metabolism of citrate until all the glucose or fructose had been metabolised. Diauxie growth was not observed but growth on glucose and fructose was much faster than on citrate. In contrast, citrate was co-metabolized with galactose or sucrose and pyruvate with glucose. When glucose was added to cells growing on citrate, glucose metabolism began immediately but inhibition of citrate utilisation did not begin for approximately 1.5 h. Growth rates of E. faecalis JH2-2 and its isogenic, catabolite derepressed mutant, strain CL4, on citrate, were 0.41 and 0.36 h(-1), respectively. The catabolite derepressed mutant was able to co-metabolise citrate and glucose at all concentrations of glucose tested (3-25 mM), while its parent, could only metabolise citrate once all the glucose had been consumed. In strains JH2-2 and E-239, the growth rate on citrate decreased as the glucose concentration increased and, in 25 mM glucose, consumption of citrate was inhibited for several hours after glucose had been consumed. These results indicate that catabolite repression by glucose and fructose occurs in enterococci.

Effect of raw-milk cheese consumption on the enterococcal flora of human feces.

Enterococci are one of the major facultative anaerobic bacterial groups that reside in the human gastrointestinal tract. In the present study, the composition of the enterococcal fecal flora in three healthy humans was analyzed before, during, and after the daily consumption of approximately 125 g of a raw-milk Cheddar-type cheese containing 3.2 x 10(4) enterococci/g of cheese. Enterococcal counts ranged between 1.4 x 10(2) and 2.5 x 10(8) CFU/g of feces and differed from subject to subject and from week to week. The cheese contained mainly Enterococcus casseliflavus and a small population of Enterococcus faecalis. Clonal relationships were determined by pulsed-field gel electrophoresis. Before and after consumption of the cheese, samples from humans contained mainly Enterococcus faecium, with some of the clones being resident. During consumption of the cheese, one particular transient clone of E. faecalis, clone Fs2, which was present in small numbers in the cheese, largely dominated the feces. Two clones of E. casseliflavus from the cheese were also found in the feces of one of the subjects during cheese consumption. These results suggest that a clone need not be present in a food in high numbers to establish itself in the intestine.

Biodiversity of the bacterial flora on the surface of a smear cheese.

The bacteria on the surface of a farmhouse smear-ripened cheese at four stages of ripening (4, 16, 23, and 37 days) from inoculated (i.e., deliberately inoculated with Brevibacterium linens BL2) and noninoculated (not deliberately inoculated with B. linens BL2) cheese were investigated. The results show that, contrary to accepted belief, B. linens is not a significant member of the surface flora of smear cheese and no microbial succession of species occurred during the ripening of the cheeses. Of 400 isolates made, 390 were lactate-utilizing coryneforms and 10 were coagulase-negative Staphylococcus spp. A detailed analysis of the coryneforms was undertaken using phenotypic analysis, molecular fingerprinting, chemotaxonomic techniques, and 16S rRNA gene sequencing. DNA banding profiles (ramdom amplified polymorphic DNA [RAPD]-PCR) of all the coryneform isolates showed large numbers of clusters. However, pulsed-field gel electrophoresis (PFGE) of the isolates from the cheeses showed that all isolates within a cluster and in many contiguous clusters were the same. The inoculated and noninoculated cheeses were dominated by single clones of novel species of Corynebacterium casei (50.2% of isolates), Corynebacterium mooreparkense (26% of isolates), and Microbacterium gubbeenense (12.8% of isolates). In addition, five of the isolates from the inoculated cheese were Corynebacterium flavescens. Thirty-seven strains were not identified but many had similar PFGE patterns, indicating that they were the same species. C. mooreparkense and C. casei grew at pH values below 4.9 in the presence of 8% NaCl, while M. gubbeenense did not grow below pH 5.8 in the presence of 5 to 10% NaCl. B. linens BL2 was not recovered from the inoculated cheese because it was inhibited by all the Staphylococcus isolates and many of the coryneforms. It was concluded that within a particular batch of cheese there was significant bacterial diversity in the microflora on the surface.