Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

Deep learning model for classifying endometrial lesions.

In Journal of translational medicine

BACKGROUND : Hysteroscopy is a commonly used technique for diagnosing endometrial lesions. It is essential to develop an objective model to aid clinicians in lesion diagnosis, as each type of lesion has a distinct treatment, and judgments of hysteroscopists are relatively subjective. This study constructs a convolutional neural network model that can automatically classify endometrial lesions using hysteroscopic images as input.

METHODS : All histopathologically confirmed endometrial lesion images were obtained from the Shengjing Hospital of China Medical University, including endometrial hyperplasia without atypia, atypical hyperplasia, endometrial cancer, endometrial polyps, and submucous myomas. The study included 1851 images from 454 patients. After the images were preprocessed (histogram equalization, addition of noise, rotations, and flips), a training set of 6478 images was input into a tuned VGGNet-16 model; 250 images were used as the test set to evaluate the model's performance. Thereafter, we compared the model's results with the diagnosis of gynecologists.

RESULTS : The overall accuracy of the VGGNet-16 model in classifying endometrial lesions is 80.8%. Its sensitivity to endometrial hyperplasia without atypia, atypical hyperplasia, endometrial cancer, endometrial polyp, and submucous myoma is 84.0%, 68.0%, 78.0%, 94.0%, and 80.0%, respectively; for these diagnoses, the model's specificity is 92.5%, 95.5%, 96.5%, 95.0%, and 96.5%, respectively. When classifying lesions as benign or as premalignant/malignant, the VGGNet-16 model's accuracy, sensitivity, and specificity are 90.8%, 83.0%, and 96.0%, respectively. The diagnostic performance of the VGGNet-16 model is slightly better than that of the three gynecologists in both classification tasks. With the aid of the model, the overall accuracy of the diagnosis of endometrial lesions by gynecologists can be improved.

CONCLUSIONS : The VGGNet-16 model performs well in classifying endometrial lesions from hysteroscopic images and can provide objective diagnostic evidence for hysteroscopists.

Zhang YunZheng, Wang ZiHao, Zhang Jin, Wang CuiCui, Wang YuShan, Chen Hao, Shan LuHe, Huo JiaNing, Gu JiaHui, Ma Xiaoxin

2021-Jan-06

Computer-aided diagnosis, Convolutional neural network, Endometrial lesion, Hysteroscopy, VGGNet

Ophthalmology Ophthalmology

An innovative strategy for standardized, structured, and interoperable results in ophthalmic examinations.

In BMC medical informatics and decision making ; h5-index 38.0

BACKGROUND : Although ophthalmic devices have made remarkable progress and are widely used, most lack standardization of both image review and results reporting systems, making interoperability unachievable. We developed and validated new software for extracting, transforming, and storing information from report images produced by ophthalmic examination devices to generate standardized, structured, and interoperable information to assist ophthalmologists in eye clinics.

RESULTS : We selected report images derived from optical coherence tomography (OCT). The new software consists of three parts: (1) The Area Explorer, which determines whether the designated area in the configuration file contains numeric values or tomographic images; (2) The Value Reader, which converts images to text according to ophthalmic measurements; and (3) The Finding Classifier, which classifies pathologic findings from tomographic images included in the report. After assessment of Value Reader accuracy by human experts, all report images were converted and stored in a database. We applied the Value Reader, which achieved 99.67% accuracy, to a total of 433,175 OCT report images acquired in a single tertiary hospital from 07/04/2006 to 08/31/2019. The Finding Classifier provided pathologic findings (e.g., macular edema and subretinal fluid) and disease activity. Patient longitudinal data could be easily reviewed to document changes in measurements over time. The final results were loaded into a common data model (CDM), and the cropped tomographic images were loaded into the Picture Archive Communication System.

CONCLUSIONS : The newly developed software extracts valuable information from OCT images and may be extended to other types of report image files produced by medical devices. Furthermore, powerful databases such as the CDM may be implemented or augmented by adding the information captured through our program.

Mun Yongseok, Kim Jooyoung, Noh Kyoung Jin, Lee Soochahn, Kim Seok, Yi Soyoung, Park Kyu Hyung, Yoo Sooyoung, Chang Dong Jin, Park Sang Jun

2021-Jan-06

Deep learning, Optical character recognition, Optical coherence tomography, Text detection

oncology Oncology

Generating real-world evidence from unstructured clinical notes to examine clinical utility of genetic tests: use case in BRCAness.

In BMC medical informatics and decision making ; h5-index 38.0

BACKGROUND : Next-generation sequencing provides comprehensive information about individuals' genetic makeup and is commonplace in oncology clinical practice. However, the utility of genetic information in the clinical decision-making process has not been examined extensively from a real-world, data-driven perspective. Through mining real-world data (RWD) from clinical notes, we could extract patients' genetic information and further associate treatment decisions with genetic information.

METHODS : We proposed a real-world evidence (RWE) study framework that incorporates context-based natural language processing (NLP) methods and data quality examination before final association analysis. The framework was demonstrated in a Foundation-tested women cancer cohort (N = 196). Upon retrieval of patients' genetic information using NLP system, we assessed the completeness of genetic data captured in unstructured clinical notes according to a genetic data-model. We examined the distribution of different topics regarding BRCA1/2 throughout patients' treatment process, and then analyzed the association between BRCA1/2 mutation status and the discussion/prescription of targeted therapy.

RESULTS : We identified seven topics in the clinical context of genetic mentions including: Information, Evaluation, Insurance, Order, Negative, Positive, and Variants of unknown significance. Our rule-based system achieved a precision of 0.87, recall of 0.93 and F-measure of 0.91. Our machine learning system achieved a precision of 0.901, recall of 0.899 and F-measure of 0.9 for four-topic classification and a precision of 0.833, recall of 0.823 and F-measure of 0.82 for seven-topic classification. We found in result-containing sentences, the capture of BRCA1/2 mutation information was 75%, but detailed variant information (e.g. variant types) is largely missing. Using cleaned RWD, significant associations were found between BRCA1/2 positive mutation and targeted therapies.

CONCLUSIONS : In conclusion, we demonstrated a framework to generate RWE using RWD from different clinical sources. Rule-based NLP system achieved the best performance for resolving contextual variability when extracting RWD from unstructured clinical notes. Data quality issues such as incompleteness and discrepancies exist thus manual data cleaning is needed before further analysis can be performed. Finally, we were able to use cleaned RWD to evaluate the real-world utility of genetic information to initiate a prescription of targeted therapy.

Zhao Yiqing, Weroha Saravut J, Goode Ellen L, Liu Hongfang, Wang Chen

2021-Jan-06

BRCA1/2, Electronic health records, Natural language processing, PARP inhibitor, Precision medicine, Real-world evidence

General General

Evaluating performance of health care facilities at meeting HIV-indicator reporting requirements in Kenya: an application of K-means clustering algorithm.

In BMC medical informatics and decision making ; h5-index 38.0

BACKGROUND : The ability to report complete, accurate and timely data by HIV care providers and other entities is a key aspect in monitoring trends in HIV prevention, treatment and care, hence contributing to its eradication. In many low-middle-income-countries (LMICs), aggregate HIV data reporting is done through the District Health Information Software 2 (DHIS2). Nevertheless, despite a long-standing requirement to report HIV-indicator data to DHIS2 in LMICs, few rigorous evaluations exist to evaluate adequacy of health facility reporting at meeting completeness and timeliness requirements over time. The aim of this study is to conduct a comprehensive assessment of the reporting status for HIV-indicators, from the time of DHIS2 implementation, using Kenya as a case study.

METHODS : A retrospective observational study was conducted to assess reporting performance of health facilities providing any of the HIV services in all 47 counties in Kenya between 2011 and 2018. Using data extracted from DHIS2, K-means clustering algorithm was used to identify homogeneous groups of health facilities based on their performance in meeting timeliness and completeness facility reporting requirements for each of the six programmatic areas. Average silhouette coefficient was used in measuring the quality of the selected clusters.

RESULTS : Based on percentage average facility reporting completeness and timeliness, four homogeneous groups of facilities were identified namely: best performers, average performers, poor performers and outlier performers. Apart from blood safety reports, a distinct pattern was observed in five of the remaining reports, with the proportion of best performing facilities increasing and the proportion of poor performing facilities decreasing over time. However, between 2016 and 2018, the proportion of best performers declined in some of the programmatic areas. Over the study period, no distinct pattern or trend in proportion changes was observed among facilities in the average and outlier groups.

CONCLUSIONS : The identified clusters revealed general improvements in reporting performance in the various reporting areas over time, but with noticeable decrease in some areas between 2016 and 2018. This signifies the need for continuous performance monitoring with possible integration of machine learning and visualization approaches into national HIV reporting systems.

Gesicho Milka Bochere, Were Martin Chieng, Babic Ankica

2021-Jan-06

Completeness, DHIS2, K-means clustering, Performance, Timeliness

Public Health Public Health

"AI's gonna have an impact on everything in society, so it has to have an impact on public health": a fundamental qualitative descriptive study of the implications of artificial intelligence for public health.

In BMC public health ; h5-index 82.0

BACKGROUND : Our objective was to determine the impacts of artificial intelligence (AI) on public health practice.

METHODS : We used a fundamental qualitative descriptive study design, enrolling 15 experts in public health and AI from June 2018 until July 2019 who worked in North America and Asia. We conducted in-depth semi-structured interviews, iteratively coded the resulting transcripts, and analyzed the results thematically.

RESULTS : We developed 137 codes, from which nine themes emerged. The themes included opportunities such as leveraging big data and improving interventions; barriers to adoption such as confusion regarding AI's applicability, limited capacity, and poor data quality; and risks such as propagation of bias, exacerbation of inequity, hype, and poor regulation.

CONCLUSIONS : Experts are cautiously optimistic about AI's impacts on public health practice, particularly for improving disease surveillance. However, they perceived substantial barriers, such as a lack of available expertise, and risks, including inadequate regulation. Therefore, investment and research into AI for public health practice would likely be beneficial. However, increased access to high-quality data, research and education regarding the limitations of AI, and development of rigorous regulation are necessary to realize these benefits.

Morgenstern Jason D, Rosella Laura C, Daley Mark J, Goel Vivek, Sch√ľnemann Holger J, Piggott Thomas

2021-Jan-06

Big data, Community medicine, Epidemiology, Machine learning, Population health, Preventive medicine, Qualitative

General General

Melanoma diagnosis using deep learning techniques on dermatoscopic images.

In BMC medical imaging

BACKGROUND : Melanoma has become more widespread over the past 30 years and early detection is a major factor in reducing mortality rates associated with this type of skin cancer. Therefore, having access to an automatic, reliable system that is able to detect the presence of melanoma via a dermatoscopic image of lesions and/or skin pigmentation can be a very useful tool in the area of medical diagnosis.

METHODS : Among state-of-the-art methods used for automated or computer assisted medical diagnosis, attention should be drawn to Deep Learning based on Convolutional Neural Networks, wherewith segmentation, classification and detection systems for several diseases have been implemented. The method proposed in this paper involves an initial stage that automatically crops the region of interest within a dermatoscopic image using the Mask and Region-based Convolutional Neural Network technique, and a second stage based on a ResNet152 structure, which classifies lesions as either "benign" or "malignant".

RESULTS : Training, validation and testing of the proposed model was carried out using the database associated to the challenge set out at the 2017 International Symposium on Biomedical Imaging. On the test data set, the proposed model achieves an increase in accuracy and balanced accuracy of 3.66% and 9.96%, respectively, with respect to the best accuracy and the best sensitivity/specificity ratio reported to date for melanoma detection in this challenge. Additionally, unlike previous models, the specificity and sensitivity achieve a high score (greater than 0.8) simultaneously, which indicates that the model is good for accurate discrimination between benign and malignant lesion, not biased towards any of those classes.

CONCLUSIONS : The results achieved with the proposed model suggest a significant improvement over the results obtained in the state of the art as far as performance of skin lesion classifiers (malignant/benign) is concerned.

Jojoa Acosta Mario Fernando, Caballero Tovar Liesle Yail, Garcia-Zapirain Maria Begonya, Percybrooks Winston Spencer

2021-Jan-06

Convolutional neural network, Deep learning, Mask R_CNN, Object classification, Object detection, Transfer learning