Automated Machine Learning Approach Research Articles

Machine Learning and Expert Judgement: Analyzing Emerging Topics in Accounting and Finance Research in the Asia–Pacific

In this paper, we focus on the question to what extent machine learning (ML) tools can be used to support systematic literature reviews. We apply a ML approach for topic detection to analyze emerging topics in the literature—our context is accounting and finance research in the Asia–Pacific region. To evaluate the robustness of the approach, we compare findings from the automated ML approach with the results from a manual analysis of the literature. The automated approach uses a keyword algorithm detection mechanism whereby the manual analysis uses common techniques for qualitative data analysis, that is, triangulation between researchers (expert judgement). From our paper, we conclude that both methods have strengths and weaknesses. The automated analysis works well for large corpora of text and provides a very standardized and non‐biased way of analyzing the literature. However, the human researcher is potentially better equipped to evaluate current issues and future trends in the literature. Overall, the best results might be achieved when a variety of tools are used together.

Automated machine learning approaches for humpback whale vocalization classification

The vocalization behavior of humpback whales was previously studied and mapped over instantaneous wide areas of the Gulf of Maine (GOM), spanning more than 100 km in diameter, using the passive ocean acoustic waveguide remote sensing technique during their Fall feeding season. Acoustic signals were received on a 160-element hydrophone array system where beamforming was employed to significantly improve signal-to-noise ratio of the received vocalizations. The humpback whale vocalizations can be divided into two classes, song and non-song calls. Song vocalizations are composed of repeatable set of phrases with consistently short inter-pulse intervals. The non-song vocalizations, such as ‘bow-shaped’ and ‘downsweep’ moans, have large and highly variable inter-pulse intervals and no repeatable pattern. Here we employ machine learning approaches to classify humpback whale vocalizations into song and non-song calls. Preprocessing methods including frequency filtering, wavelet denoising, beamforming, and spectral smoothing are applied. Several automated classification methods including Support Vector Machines, Gaussian Naive Bayes, and Neural Networks are explored. Implementation of these algorithms on the GOM dataset results in over 88% classification accuracy implying the machine learning approaches can be used in field studies for real-time classification.

Otoscopic diagnosis using computer vision: An automated machine learning approach.

Access to otolaryngology is limited by lengthy wait lists and lack of specialists, especially in rural and remote areas. The objective of this study was to use an automated machine learning approach to build a computer vision algorithm for otoscopic diagnosis capable of greater accuracy than trained physicians. This algorithm could be used by primary care providers to facilitate timely referral, triage, and effective treatment. Otoscopic images were obtained from Google Images (Google Inc., Mountain View, CA), from open access repositories, and within otolaryngology clinics associated with our institution. After preprocessing, 1,366 unique images were uploaded to the Google Cloud Vision AutoML platform (Google Inc.) and annotated with one or more of 14 otologic diagnoses. A consensus set of labels for each otoscopic image was attained, and a multilabel classifier architecture algorithm was trained. The performance of the algorithm on an 89-image test set was compared to the performance of physicians from pediatrics, emergency medicine, otolaryngology, and family medicine. For all diagnoses combined, the average precision (positive predictive value) of the algorithm was 90.9%, and the average recall (sensitivity) was 86.1%. The algorithm made 79 correct diagnoses with an accuracy of 88.7%. The average physician accuracy was 58.9%. We have created a computer vision algorithm using automated machine learning that on average rivals the accuracy of the physicians we tested. Fourteen different otologic diagnoses were analyzed. The field of medicine will be changed dramatically by artificial intelligence within the next few decades, and physicians of all specialties must be prepared to guide that process. NA Laryngoscope, 130:1408-1413, 2020.

Automating Ischemic Stroke Subtype Classification Using Machine Learning and Natural Language Processing

Objective: The manual adjudication of disease classification is time-consuming, error-prone, and limits scaling to large datasets. In ischemic stroke (IS), subtype classification is critical for management and outcome prediction. This study sought to use natural language processing of electronic health records (EHR) combined with machine learning methods to automate IS subtyping. Methods: Among IS patients from an observational registry with TOAST subtyping adjudicated by board-certified vascular neurologists, we analyzed unstructured text-based EHR data including neurology progress notes and neuroradiology reports using natural language processing. We performed several feature selection methods to reduce the high dimensionality of the features and 5-fold cross validation to test generalizability of our methods and minimize overfitting. We used several machine learning methods and calculated the kappa values for agreement between each machine learning approach to manual adjudication. We then performed a blinded testing of the best algorithm against a held-out subset of 50 cases. Results: Compared to manual classification, the best machine-based classification achieved a kappa of .25 using radiology reports alone, .57 using progress notes alone, and .57 using combined data. Kappa values varied by subtype being highest for cardioembolic (.64) and lowest for cryptogenic cases (.47). In the held-out test subset, machine-based classification agreed with rater classification in 40 of 50 cases (kappa .72). Conclusions: Automated machine learning approaches using textual data from the EHR shows agreement with manual TOAST classification. The automated pipeline, if externally validated, could enable large-scale stroke epidemiology research.

The dynamic conformational landscape of the protein methyltransferase SETD8.

Elucidating the conformational heterogeneity of proteins is essential for understanding protein function and developing exogenous ligands. With the rapid development of experimental and computational methods, it is of great interest to integrate these approaches to illuminate the conformational landscapes of target proteins. SETD8 is a protein lysine methyltransferase (PKMT), which functions in vivo via the methylation of histone and nonhistone targets. Utilizing covalent inhibitors and depleting native ligands to trap hidden conformational states, we obtained diverse X-ray structures of SETD8. These structures were used to seed distributed atomistic molecular dynamics simulations that generated a total of six milliseconds of trajectory data. Markov state models, built via an automated machine learning approach and corroborated experimentally, reveal how slow conformational motions and conformational states are relevant to catalysis. These findings provide molecular insight on enzymatic catalysis and allosteric mechanisms of a PKMT via its detailed conformational landscape.

DeephESC 2.0: Deep Generative Multi Adversarial Networks for improving the classification of hESC.

Human embryonic stem cells (hESC), derived from the blastocysts, provide unique cellular models for numerous potential applications. They have great promise in the treatment of diseases such as Parkinson’s, Huntington’s, diabetes mellitus, etc. hESC are a reliable developmental model for early embryonic growth because of their ability to divide indefinitely (pluripotency), and differentiate, or functionally change, into any adult cell type. Their adaptation to toxicological studies is particularly attractive as pluripotent stem cells can be used to model various stages of prenatal development. Automated detection and classification of human embryonic stem cell in videos is of great interest among biologists for quantified analysis of various states of hESC in experimental work. Currently video annotation is done by hand, a process which is very time consuming and exhaustive. To solve this problem, this paper introduces DeephESC 2.0 an automated machine learning approach consisting of two parts: (a) Generative Multi Adversarial Networks (GMAN) for generating synthetic images of hESC, (b) a hierarchical classification system consisting of Convolution Neural Networks (CNN) and Triplet CNNs to classify phase contrast hESC images into six different classes namely: Cell clusters, Debris, Unattached cells, Attached cells, Dynamically Blebbing cells and Apoptically Blebbing cells. The approach is totally non-invasive and does not require any chemical or staining of hESC. DeephESC 2.0 is able to classify hESC images with an accuracy of 93.23% out performing state-of-the-art approaches by at least 20%. Furthermore, DeephESC 2.0 is able to generate large number of synthetic images which can be used for augmenting the dataset. Experimental results show that training DeephESC 2.0 exclusively on a large amount of synthetic images helps to improve the performance of the classifier on original images from 93.23% to 94.46%. This paper also evaluates the quality of the generated synthetic images using the Structural SIMilarity (SSIM) index, Peak Signal to Noise ratio (PSNR) and statistical p-value metrics and compares them with state-of-the-art approaches for generating synthetic images. DeephESC 2.0 saves hundreds of hours of manual labor which would otherwise be spent on manually/semi-manually annotating more and more videos.

04:12 PM Abstract No. 41 Automated pre-procedure telephone calls performed by a machine learning agent: a novel artificial intelligence technique for risk minimization

Appropriate pre-procedure communication is critical to a successful interventional radiology procedure. In the traditional approach, a healthcare professional contacts the patient to discuss issues such as pre-procedure fasting, blood thinners, and medication/contrast allergies. We use machine learning to create a virtual agent that can perform these tasks in an automated way. We have developed a conversational agent that interacts with patients via telephone. It follows a pre-procedure checklist to address issues such as allergies and blood thinner use. The patient is then able to freely query the agent about their procedure and receive intelligent responses, including information related to pain control, procedure length, and directions to the procedure area. This conversational agent was built using the Google Phone Gateway technology within the Google Cloud product suite. A Python script running on a Heroku server performs various operations, including constructing responses to user input and sending a summary of each interaction to a central database via email. The conversational agent was able to effectively interact with users via telephone and compile relevant pre-procedure information. The ability to understand conversational user queries and respond appropriately improved significantly with larger amounts of training data. Optimal performance was seen with a classification threshold of 0.3 and a combined rule-based/machine learning technique. Various pre-procedure issues create substantial risk of procedure cancellation or patient harm, such as eating prior to a procedure or an unidentified contrast allergy. Although screening for these issues is generally performed by human experts, weaknesses of this approach include labor costs, lack of standardization, and the potential for human error. We present an automated machine learning approach that offers an inexpensive, systematic tool for pre-procedure planning. We believe optimal results will be seen in a strategy that combines both human evaluation and assessment by a standardized machine learning agent.

Glacier Facies Mapping Using a Machine-Learning Algorithm: The Parlung Zangbo Basin Case Study

Glaciers in the Tibetan Plateau are an important indicator of climate change. Automatic glacier facies mapping utilizing remote sensing data is challenging due to the spectral similarity of supraglacial debris and the adjacent bedrock. Most of the available glacier datasets do not provide the boundary of clean ice and debris-covered glacier facies, while debris-covered glacier facies play a key role in mass balance research. The aim of this study was to develop an automatic algorithm to distinguish ice cover types based on multi-temporal satellite data, and the algorithm was implemented in a subregion of the Parlung Zangbo basin in the southeastern Tibetan Plateau. The classification method was built upon an automated machine learning approach: Random Forest in combination with the analysis of topographic and textural features based on Landsat-8 imagery and multiple digital elevation model (DEM) data. Very high spatial resolution Gao Fen-1 (GF-1) Panchromatic and Multi-Spectral (PMS) imagery was used to select training samples and validate the classification results. In this study, all of the land cover types were classified with overall good performance using the proposed method. The results indicated that fully debris-covered glaciers accounted for approximately 20.7% of the total glacier area in this region and were mainly distributed at elevations between 4600 m and 4800 m above sea level (a.s.l.). Additionally, an analysis of the results clearly revealed that the proportion of small size glaciers (<1 km2) were 88.3% distributed at lower elevations compared to larger size glaciers (≥1 km2). In addition, the majority of glaciers (both in terms of glacier number and area) were characterized by a mean slope ranging between 20° and 30°, and 42.1% of glaciers had a northeast and north orientation in the Parlung Zangbo basin.

A Deep Learning Approach to Vascular Structure Segmentation in Dermoscopy Colour Images.

Background Atypical vascular pattern is one of the most important features by differentiating between benign and malignant pigmented skin lesions. Detection and analysis of vascular structures is a necessary initial step for skin mole assessment; it is a prerequisite step to provide an accurate outcome for the widely used 7-point checklist diagnostic algorithm. Methods In this research we present a fully automated machine learning approach for segmenting vascular structures in dermoscopy colour images. The U-Net architecture is based on convolutional networks and designed for fast and precise segmentation of images. After preprocessing the images are randomly divided into 146516 patches of 64 × 64 pixels each. Results On the independent validation dataset including 74 images our implemented method showed high segmentation accuracy. For the U-Net convolutional neural network, an average DSC of 0.84, sensitivity 0.85, and specificity 0.81 has been achieved. Conclusion Vascular structures due to small size and similarity to other local structures create enormous difficulties during the segmentation and assessment process. The use of advanced segmentation methods like deep learning, especially convolutional neural networks, has the potential to improve the accuracy of advanced local structure detection.

Abstract A56: Development of a cloud-based machine learning system (CLOBNET) to predict platinum resistance in high-grade serous ovarian cancer

Abstract Platinum resistance is a major therapeutic obstacle in the treatment of patients with high-grade serous ovarian cancer (HGSOC), and predicting poor response to first-line therapies would allow for more rapid implementation of alternative treatment approaches. At diagnosis, clinicians often have limited access to heavily processed genomics data to predict treatment responses. The use of rich clinical data accumulated into electronic health records (EHRs) could have features able to discriminate nonresponsive patients, but their use is currently hampered by multiple manual steps and lack of automated analytic systems. We have developed a cloud-based machine learning system (CLOBNET), with streamlined collection and real-time analysis of rich clinical data from EHRs, and use the CLOBNET to predict platinum resistance in patients with HGSOC. This study is a part of research program HERCULES, and The Ethics Committee of The Hospital District of Southwest Finland has approved the study. The main system is based on a cloud server inside Turku University Hospital’s cloud server and can be accessed via encrypted tunnel from anywhere outside. Data are collected automatically by the hospital’s clinical research center from various different live databases and duplicated to a single SQL database inside the cloud server. This reduces labor, keeps information up to date, and reduces the possibility of human error in data input. Data are analyzed by server back-end Python scripts or by web-based front-end, depending on demand of the algorithm. Data and other information are visualized as a dynamic web page constructed with PHP, JavaScript libraries such as Chart.js, and styled with Bootstrap. Every component used to create the system is open source, so extensive documentation and help can be found online from developer communities such as Stack Overflow or GitHub. Data protection and patient privacy are strictly protected while patient data are accessed over the internet by using several steps; CLOBNET can be accessed only through Turku University Hospital’s proxy server using double SSH tunnel and public key authentication and every authentication step and database access also require strong passwords. Encryption used in our server traffic is on par with online banking encryption. Clinical data of 235 patients with informed consent were collected during a prospective study from Turku University Hospital during 2009 to 2017. We next used CLOBNET to predict the response to platinum-based therapies using a linear normal-based classifier algorithm created with Matlab. Feature selection for the model was done using branch and bound algorithm and features with over 20% missing values were omitted. With a subpopulation of 56 patients who had had one operation and thus had a defined tumor dissemination, a model was trained to differentiate between complete response and progressive disease. With AUC of 0,80 our model could predict complete response with 94.1% sensitivity and 45.5% specificity. Using a 40% missing value cut point, 28 patients remained for testing a model trained to identify patients with platinum resistance (platinum-free interval of less than six months). The model achieved sensitivity of 84.2% and specificity of 77.8% with AUC of 0.86. We have herein shown the viability of accessing clinical data from live EHRs and using machine learning to analyze it over traditionally restricting barriers caused by manual steps and interfaces between different organizations and systems. Even with a limited dataset we can train a linear model to predict primary progressive disease and platinum resistance in patients with HGSOC. Thus, automated machine learning approaches, such as CLOBNET, show promise in becoming clinical tools in predicting challenging outcomes, such as platinum resistance in HGSOC. Citation Format: Veli-Matti Isoviita, Liina Salminen, Jimmy Azar, Johanna Hynninen, Rainer Lehtonen, Pia Röring, Seija Grénman, Anniina Färkkilä, Sampsa Hautaniemi. Development of a cloud-based machine learning system (CLOBNET) to predict platinum resistance in high-grade serous ovarian cancer. [abstract]. In: Proceedings of the AACR Conference: Addressing Critical Questions in Ovarian Cancer Research and Treatment; Oct 1-4, 2017; Pittsburgh, PA. Philadelphia (PA): AACR; Clin Cancer Res 2018;24(15_Suppl):Abstract nr A56.

Production of Multi-Features Driven Nationwide Vegetation Physiognomic Map and Comparison to MODIS Land Cover Type Product

Irrespective of several attempts to land use/cover mapping at local, regional, or global scales, mapping of vegetation physiognomic types is limited and challenging. The main objective of the research is to produce an accurate nationwide vegetation physiognomic map by using automated machine learning approach with the support of reference data. A time-series of the multi-spectral and multi-indices data derived from Moderate Resolution Imaging Spectroradiometer (MODIS) were exploited along with the land-surface slope data. Reliable reference data of the vegetation physiognomic types were prepared by refining the existing vegetation survey data available in the country. The Random Forests based mapping framework adopted in the research showed high performance (Overall accuracy = 0.82, Kappa coefficient = 0.79) using 148 optimum number of features out of 231 featured used. A nationwide vegetation physiognomic map of year 2013 was produced in the research. The resulted map was compared to the existing MODIS Land Cover Type (MCD12Q1) product of year 2013. A huge difference was found between two maps. Validation with the reference data showed that the MCD12Q1 product did not work satisfactorily in Japan. The outcome of the research highlights the possibility of improving the accuracy of the MCD12Q1 product with special focus on reference data.

Objective detection of apoptosis in rat renal tissue sections using light microscopy and free image analysis software with subsequent machine learning: Detection of apoptosis in renal tissue.

The current study proposes an automated machine learning approach for the quantification of cells in cell death pathways according to DNA fragmentation. A total of 17 images of kidney histological slide samples from male Wistar rats were used. The slides were photographed using an Axio Zeiss Vert.A1 microscope with a 40x objective lens coupled with an Axio Cam MRC Zeiss camera and Zen 2012 software. The images were analyzed using CellProfiler (version 2.1.1) and CellProfiler Analyst open-source software. Out of the 10,378 objects, 4970 (47,9%) were identified as TUNEL positive, and 5408 (52,1%) were identified as TUNEL negative. On average, the sensitivity and specificity values of the machine learning approach were 0.80 and 0.77, respectively. Image cytometry provides a quantitative analytical alternative to the more traditional qualitative methods more commonly used in studies.

Challenges and practical approaches with word sense disambiguation of acronyms and abbreviations in the clinical domain.

ObjectivesAlthough acronyms and abbreviations in clinical text are used widely on a daily basis, relatively little research has focused upon word sense disambiguation (WSD) of acronyms and abbreviations in the healthcare domain. Since clinical notes have distinctive characteristics, it is unclear whether techniques effective for acronym and abbreviation WSD from biomedical literature are sufficient.MethodsThe authors discuss feature selection for automated techniques and challenges with WSD of acronyms and abbreviations in the clinical domain.ResultsThere are significant challenges associated with the informal nature of clinical text, such as typographical errors and incomplete sentences; difficulty with insufficient clinical resources, such as clinical sense inventories; and obstacles with privacy and security for conducting research with clinical text. Although we anticipated that using sophisticated techniques, such as biomedical terminologies, semantic types, part-of-speech, and language modeling, would be needed for feature selection with automated machine learning approaches, we found instead that simple techniques, such as bag-of-words, were quite effective in many cases. Factors, such as majority sense prevalence and the degree of separateness between sense meanings, were also important considerations.ConclusionsThe first lesson is that a comprehensive understanding of the unique characteristics of clinical text is important for automatic acronym and abbreviation WSD. The second lesson learned is that investigators may find that using simple approaches is an effective starting point for these tasks. Finally, similar to other WSD tasks, an understanding of baseline majority sense rates and separateness between senses is important. Further studies and practical solutions are needed to better address these issues.