Deep Learning Provides Rapid Screen for Breast Cancer Metastasis with Sentinel Lymph Nodes.

OBJECTIVE: Deep learning has been shown to be useful in detecting breast cancer metastases by analyzing whole slide images (WSI) of sentinel lymph nodes; however, it requires extensive analysis of all the lymph node slides. Our deep learning study attempts to provide a rapid screen for metastasis by analyzing only a small set of image patches to detect changes in tumor environment. METHODS: We designed a convolutional neural network to build a diagnostic model for metastasis detection. We obtained WSIs of Hematoxylin and Eosin-stained slides from 34 cases with equal distribution in positive/negative categories. Two WSIs were selected from each case for a total of 69 WSIs. From each WSI, 40 image patches (100x100 pixels) were obtained to yield 2720 image patches, from which 2160 (79%) were used for training, 240 (9%) for validation, and 320 (12%) for testing. Interobserver variation was also examined among 3 users. RESULTS: The test results showed excellent diagnostic results: accuracy (91.15%), sensitivity (77.92%), and specificity (92.09%). No significant variation in results was observed among the 3 observers. CONCLUSION: This preliminary study provided a proof of concept for conducting a rapid screen for metastasis rather than an exhaustive search for tumors in all fields of all sentinel lymph nodes.

Distortions but not Definitive Peaks by Capillary Serum Protein Electrophoresis Indicate Monoclonal Gammopathy of Undetermined Significance rather than Multiple Myeloma.

OBJECTIVE: Capillary electrophoresis of serum proteins demonstrates occasional distortions. Distortions or peaks in the gamma, beta, and alpha-2 zones may represent monoclonal gammopathy. In this study, we investigated if such distortions are associated with monoclonal gammopathy of undetermined significance (MGUS) or multiple myeloma. METHODS: Consecutive serum protein electrophoresis results were reviewed and immunofixation studies were recommended on specimens exhibiting distortions or distinct peaks in the gamma, beta or alpha-2 zones. RESULTS AND DISCUSSION: Of the 471 cases, we observed distortions in 101 cases. In the immunofixation studies, 17.8% of cases had a diagnosis of MGUS, but none contained multiple myeloma. CONCLUSIONS: We conclude that distortions in serum capillary electrophoresis may be associated with MGUS, but not multiple myeloma.

Automated Diagnosis of Lymphoma with Digital Pathology Images Using Deep Learning.

Recent studies have shown promising results in using Deep Learning to detect malignancy in whole slide imaging, however, they were limited to just predicting a positive or negative finding for a specific neoplasm. We attempted to use Deep Learning with a convolutional neural network (CNN) algorithm to build a lymphoma diagnostic model for four diagnostic categories: (1) benign lymph node, (2) diffuse large B-cell lymphoma, (3) Burkitt lymphoma, and (4) small lymphocytic lymphoma. Our software was written in Python language. We obtained digital whole-slide images of Hematoxylin and Eosin stained slides of 128 cases including 32 cases for each diagnostic category. Four sets of 5 representative images, 40x40 pixels in dimension, were taken for each case. A total of 2,560 images were obtained from which 1,856 were used for training, 464 for validation, and 240 for testing. For each test set of 5 images, the predicted diagnosis was combined from the prediction of five images. The test results showed excellent diagnostic accuracy at 95% for image-by-image prediction and at 100% for set-by-set prediction. This preliminary study provided a proof of concept for incorporating automated lymphoma diagnostic screen into future pathology work-flow to augment the pathologists' productivity.

Proteomics Analysis of FLT3-ITD Mutation in Acute Myeloid Leukemia Using Deep Learning Neural Network.

Deep Learning can significantly benefit cancer proteomics and genomics. In this study, we attempted to determine a set of critical proteins that were associated with the FLT3-ITD mutation in newly-diagnosed acute myeloid leukemia patients. A Deep Learning network consisting of autoencoders formed a hierarchical model from which high-level features were extracted without labeled training data. Dimensional reduction reduced the number of critical proteins from 231 to 20. Deep Learning found an excellent correlation between FLT3-ITD mutation with the levels of these 20 critical proteins (accuracy 97%, sensitivity 90%, and specificity 100%). Our Deep Learning network could hone in on 20 proteins with the strongest association with FLT3-ITD. The results of this study allow for a novel approach to determine critical protein pathways in the FLT3-ITD mutation, and provide proof-of-concept for an accurate approach to model big data in cancer proteomics and genomics.

Utility of VerifyNow for Point-of-Care Identification of an Aspirin Effect Prior to Emergency Cardiac Surgery.

BACKGROUND: Patients with cardiovascular disease are frequently on aspirin, which may place them at risk for bleeding during surgical procedures. The utility of the VerifyNow test to rapidly identify an aspirin effect and predict bleeding risk prior to cardiac surgery was explored. METHODS: A retrospective study was performed of patients on a clinical pathology consultation service that provides laboratory and transfusion support for patients undergoing major cardiac surgery. Patients who had VerifyNow testing for aspirin effect were selected. RESULTS: A total of 88 patients had VerifyNow aspirin testing during the study period. The VerifyNow test correctly identified 52/63 (82.5%) patients with documented aspirin use, and missed 11/63 (17.5%) of aspirin users. Light transmission aggregometry (LTA) showed an aspirin effect in the majority of aspirin users missed by the VerifyNow assay. Moderate correlations were found between LTA and VerifyNow. Low aspirin reaction units were not associated with significant bleeding in these cardiac surgery patients. CONCLUSIONS: We propose the VerifyNow assay for point-of-care identification of aspirin effect prior to emergency surgeries.

Clot Strength as Measured by Thrombelastography Correlates with Platelet Reactivity in Stroke Patients.

OBJECTIVE: Platelet reactivity may be important in the management of patients with stroke. However, degree of platelet reactivity has not been correlated with Thrombelastography (TEG(®)) parameters in stroke. We sought to detect a correlation between TEG(®) values and clot platelet reactivity in ex vivo clots of stroke patients. METHODS: We collected venous blood from 40 patients with stroke. TEG(®) measurements were carried out and residual clots were fixed in 10% formalin immediately following completion of TEG(®). The formalin specimens were embedded in paraffin blocks, cut at 4 micrometers, and stained with CD 61 (immunohistochemical stain used to detect platelets) with appropriate controls. Under light microscopy, three pathologists blinded to TEG(®) results independently graded CD61 intensity (how aggregated/intense the CD61 stained) into a low and high group, as a proposed measurement representing the platelet reactivity of the clot. We compared pre-tPA-TEG(®) values among groups with different CD 61 intensities. RESULTS: After adjusting for confounding factors, we found statistically significant correlation between CD61 staining and several TEG(®) parameters (Delta and CD61 staining intensity (p=0.047); Angle and CD61 staining intensity grade (p=0.04); and G and CD61 staining intensity grade (p=0.04)). CONCLUSIONS: Clot strength on TEG(®) as measured by Delta, Angle, and G correlates with a clot with greater platelet reactivity.

Post-operative bleeding risk stratification in cardiac pulmonary bypass patients using artificial neural network.

The prediction of bleeding risk in cardiopulmonary bypass (CPB) patients plays a vital role in their postoperative management. Therefore, an artificial neural network (ANN) to analyze intra-operative laboratory data to predict postoperative bleeding was set up. The JustNN software (Neural Planner Software, Cheshire, England) was used. This ANN was trained using 15 intra-operative laboratory parameters paired with one output category - risk of bleeding, defined as units of blood components transfused in 48 hours. The ANN was trained with the first 39 CPB cases. The set of input parameters for this ANN was also determined, and the ANN was validated with the next 13 cases. The set of input parameters include five components: pro-thrombin time, platelet count, thromboelastograph-reaction time, D-Dimer, and thromboelastograph-coagulation index. The validation results show 9 cases (69.2%) with exact match, 3 cases (23.1%) with one-grading difference, and 1 case (7.7%) with two-grading difference between actual blood usage versus predicted blood usage. To the best of our knowledge, ours is the first ANN developed for post-operative bleeding risk stratification of CPB patients. With promising results, we have started using this ANN to risk-stratify our CPB patients, and it has assisted us in predicting post-operative bleeding risk.

Utility of bone marrow examination for workup of fever of unknown origin in patients with HIV/AIDS.

AIMS: The utility of bone marrow aspiration and biopsy (BMAB) as a diagnostic tool in patients with HIV/AIDS and fever of unknown origin (FUO) is a subject of debate. Because highly active antiretroviral therapy has reduced incidence of opportunistic infections, it is important to reassess the efficacy of BMAB for this diagnostic purpose. To our knowledge, no such studies have been performed in Harris County which has the highest incidence of HIV in the state of Texas. METHODS: We reviewed all BMABs from patients with HIV/AIDS and FUO or persistent cytopenia(s) from 2007 to 2011. RESULTS: Of 57 evaluable patients, BMAB was positive in 24 samples by acid fast bacilli (AFB) or Gomori methenamine silver (GMS) stains (17.5%), presence of granuloma and/or lymphohistiocytic aggregates (31.6%), culture (21.0%) or a combination. Cultures demonstrated Mycobacterium avium/intracellulare (4), M tuberculosis (2), M gordonae (1), Histoplasma capsulatum (3) and Cryptococcus neoformans (2). There were three cases in which a pathogen was grown in culture but that had a negative of 'direct examination' on tissue sections (negative AFB and GMS special stains, no morphological evidence of granuloma/lymphohistiocytic infiltrates). CONCLUSIONS: This study supports the use of diagnostic BMAB as a rapid decision-making tool in patients with HIV and FUO in the proper clinical setting. BMAB demonstrated infection-related evidence prior to positive bone marrow culture in 75% of cases. Special stains and blood cultures had similar diagnostic yield, but BMAB offers faster results. Thus, this procedure assists in clinical decision making and the refinement of treatment in a more timely manner.

Inability to Measure M-Protein With Capillary Zone Electrophoresis (CAPPILLARYS 2) in Tracings With NonDiscernable Peaks.

BACKGROUND: We performed a retrospective study to illustrate the challenges with quantifying monoclonal (M)-protein in the cases of serum protein capillary zone electrophoresis (SPCZE) where no discernable peak is apparent. MATERIALS AND METHODS: We retrospectively reviewed 160 serum immunofixation electrophoresis (SIFE) that were performed at Memorial Hermann Hospital-Texas Medical Center between October 2013 and November 2013 and we identified the positive SIFE results. The corresponding SPCZE of the positive SIFE were retrieved and evaluated for the ability to quantify M-proteins in them. We define the ability to quantify M-protein as the ability for the operator of the SPCZE to identify a discernable peak and to be able to manually gate the area under the peak. RESULTS: Twenty-two cases of SIFE detected a monoclonal immunoglobulin. Of the corresponding 22 SPCZE, we could not quantify the M-protein in 6 (27.3%) of the cases. CONCLUSION: We have shown several cases where we were not able to quantify the M-protein with SPCZE. This poses a challenge in the diagnosis and management of these patients.

Interpretation of coagulation test results using a web-based reporting system.

BACKGROUND: Web-based synoptic reporting has been successfully integrated into diverse fields of pathology, improving efficiency and reducing typographic errors. Coagulation is a challenging field for practicing pathologists and pathologists-in-training alike. OBJECTIVE: To develop a Web-based program that can expedite the generation of a individualized interpretive report for a variety of coagulation tests. METHODS: We developed a Web-based synoptic reporting system composed of 119 coagulation report templates and 38 thromboelastography (TEG) report templates covering a wide range of findings. RESULTS: Our institution implemented this reporting system in July 2011; it is currently used by pathology residents and attending pathologists. Feedback from the users of these reports have been overwhelmingly positive. Surveys note the time saved and reduced errors. CONCLUSION: Our easily accessible, user-friendly, Web-based synoptic reporting system for coagulation is a valuable asset to our laboratory services.

Thromboelastography is a suboptimal test for determination of the underlying cause of bleeding associated with cardiopulmonary bypass and may not predict a hypercoagulable state.

OBJECTIVES: Patients undergoing cardiac surgery with cardiopulmonary bypass (CPB) are at risk of bleeding. The goal of this investigation was to compare thromboelastography (TEG) with standard coagulation tests (prothrombin time [PT], partial thromboplastin time [PTT], fibrinogen, and D-dimer) in patients with active bleeding. METHODS: A retrospective study of patients who underwent cardiac surgery with CPB was performed. A second analysis was performed to determine if a shortened TEG R time is associated with thrombosis. RESULTS: Paired TEG and standard coagulation tests were available from 21 bleeding patients; of the 15 patients with normal TEG values and three with a shortened R time, all had abnormalities of standard coagulation tests. Eighteen of 67 patients who underwent surgery with CPB had an episode of postoperative bleeding. The TEG R time and coagulation index, PT, and PTT collected after CPB were associated with postoperative bleeding in the univariate analysis, but only PT was independently associated with postoperative bleeding in the multivariate analysis. In the second analysis, three of 38 patients with a normal TEG and four of 43 patients with a shortened R time had a thrombotic event during hospitalization (P = 1.00). CONCLUSIONS: TEG had limited utility in identifying the underlying cause of bleeding and was not predictive of postoperative bleeding associated with cardiac surgery compared with conventional coagulation tests. A shortened TEG R time may not represent a hypercoagulable state.

How do we manage cardiopulmonary bypass coagulopathy?

BACKGROUND: Patients who undergo cardiopulmonary bypass (CPB) are at risk for coagulopathy. Suboptimal turnaround time (TAT) of laboratory coagulation testing results in empiric administration of blood products to treat massive bleeding. We describe our initiative in establishing the coagulation-based hemotherapy (CBH) service, a clinical pathology consultation service that uses rapid TAT coagulation testing and provides comprehensive assessment of bleeding in patients undergoing CPB. A transfusion algorithm that treats the underlying cause of coagulopathy was developed. STUDY DESIGN AND METHODS: The coagulation testing menu includes all aspects of coagulopathy with close proximity of the laboratory to the operating room to allow for rapid test results. The hemotherapy pathologist monitors laboratory results at several stages in surgery and uses a comprehensive algorithm to monitor a patient's hemostasis. The optimal number and type of blood products are selected when the patient is taken off CPB. RESULTS: The CBH service was consulted for 44 ventricular assist device implants, 30 heart transplants, and 31 other cardiovascular surgeries from May 2012 through November 2013. The TAT for laboratory tests was 15 minutes for complete blood count, antithrombin, and coagulation panel and 30 minutes for VerifyNow and thromboelastography, in comparison to 45 to 60 minutes in normal settings. The transfusion algorithms were used with optimal administration of blood components with preliminary data suggestive of reduced blood product usage and better patient outcomes. CONCLUSION: We described the successful introduction of a novel pathology consultation service that uses a rapid TAT coagulation testing menu with transfusion algorithms for improved management of CPB patients.

Mobile computing platform with decision support modules for hemotherapy.

OBJECTIVES: We describe the development of a mobile computing platform (MCP) with a decision support module (DSM) for patients in our coagulation-based hemotherapy service. METHODS: The core of our MCP consists of a Microsoft Excel spreadsheet template used to gather and compute data on cardiopulmonary bypass (CPB) patients intraoperatively. The DSM is embedded into the Excel file, where the user would enter in laboratory results, and through our 45 embedded algorithms, recommendations for transfusion products would be displayed in the Excel file. RESULTS: The DSM has helped decrease the time it takes to come to a transfusion recommendation, helps double-check recommendations, and is an excellent tool for teaching. Furthermore, the problems that occur with a paper system have been eliminated, and we are now able to access this information easily and reliably. CONCLUSIONS: The development and implementation of our MCP system has greatly increased the productivity and efficiency of our hemotherapy service.

Comparison between 1-needle technique versus 2-needle technique for bone marrow aspiration and biopsy procedures.

CONTEXT: Bone marrow examination is essential for diagnosis and staging of hematologic disorders. Traditionally, the bone marrow biopsy and aspirate are obtained with 2 needles at 2 separate sites. This approach is associated with significant discomfort, procedural time, and occasionally, morbidity. Although previous observations had suggested that a single-needle technique at one site is a simpler and less-painful procedure, there had been concern that the 1-needle technique may yield a suboptimal biopsy for diagnosis. OBJECTIVE: To conduct a systematic comparison of multiple parameters of bone marrow biopsy specimens obtained by the traditional 2-needle technique versus the 1-needle technique for bone marrow collection. DESIGN: We retrospectively evaluated 20 biopsy specimens obtained by each of the 2 mentioned techniques by comparing the morphologic quality of the biopsy, biopsy length, and biopsy cellularity. RESULTS: We found that the 1-needle technique yielded an adequate biopsy for diagnosis. The measured parameters of the samples obtained by the 1-needle versus 2-needle techniques were similar. CONCLUSION: This study suggests that the 1-needle technique may be preferred for bone marrow aspirate and biopsy.

A synoptic reporting system for peripheral blood smear interpretation.

Web-based synoptic reporting systems have been shown to improve efficiency, reduce turnaround time, and decrease reporting errors in reports of surgical pathology specimens and hematologic neoplasms and bone marrow. No such system has been previously described for the reporting of peripheral blood smears. We developed a Web-based synoptic reporting system composed of a knowledge base encompassing 150 peripheral blood smear report templates covering a wide range of findings. This system was used at our institution, The University of Texas Medical School at Houston, by pathology residents under the supervision of an attending pathologist to generate peripheral blood smear reports. This system was found to produce a significant reduction in typographic errors with decreased turnaround time and improved accuracy. This synoptic reporting system can help practicing pathologists and pathology trainees to draft a complete and concise report.

A teaching database for diagnosis of hematologic neoplasms using immunophenotyping by flow cytometry.

CONTEXT: In the diagnosis of lymphomas and leukemias, flow cytometry has been considered an essential addition to morphology and immunohistochemistry. The interpretation of immunophenotyping results by flow cytometry involves pattern recognition of different hematologic neoplasms that may have similar immunologic marker profiles. An important factor that creates difficulty in the interpretation process is the lack of consistency in marker expression for a particular neoplasm. For this reason, a definitive diagnostic pattern is usually not available for each specific neoplasm. Consequently, there is a need for decision support tools to assist pathology trainees in learning flow cytometric diagnosis of leukemia and lymphoma. OBJECTIVE: Development of a Web-enabled relational database integrated with decision-making tools for teaching flow cytometric diagnosis of hematologic neoplasms. DESIGN: This database has a knowledge base containing patterns of 44 markers for 37 hematologic neoplasms. We have obtained immunophenotyping data published in the scientific literature and incorporated them into a mathematical algorithm that is integrated to the database for differential diagnostic purposes. The algorithm takes into account the incidence of positive and negative expression of each marker for each disorder. RESULTS: Validation of this algorithm was performed using 92 clinical cases accumulated from 2 different medical centers. The database also incorporates the latest World Health Organization classification for hematologic neoplasms. CONCLUSIONS: The algorithm developed in this database shows significant improvement in diagnostic accuracy over our previous database prototype. This Web-based database is proposed to be a useful public resource for teaching pathology trainees flow cytometric diagnosis.

The application of data mining to flow cytometry.

Data mining is the process of automating information discovery to detect useful patterns, correlations, and trends. Existing data must be fitted into a representative model from which useful information can be derived through a variety of algorithms. The routine generation of vast amounts of data make flow cytometry a logical target for the application of data mining. This informative unit discusses the steps of the data-mining process using the immunophenotyping of hematologic neoplasms to demonstrate the application. The author describes several types of algorithms and provides a useful resource list of commercially available tools.