bims-arihec Biomed News
on Artificial intelligence in healthcare
Issue of 2020‒03‒01
eighteen papers selected by
Céline Bélanger
Cogniges Inc.


  1. Cardiovasc Res. 2020 Feb 24. pii: cvaa021. [Epub ahead of print]
    Oikonomou EK, Siddique M, Antoniades C.
      Rapid technological advances in non-invasive imaging, coupled with the availability of large data sets and the expansion of computational models and power, have revolutionized the role of imaging in medicine. Non-invasive imaging is the pillar of modern cardiovascular diagnostics, with modalities such as cardiac computed tomography (CT) now recognized as first-line options for cardiovascular risk stratification and the assessment of stable or even unstable patients. To date, cardiovascular imaging has lagged behind other fields, such as oncology, in the clinical translational of artificial intelligence (AI)-based approaches. We hereby review the current status of AI in non-invasive cardiovascular imaging, using cardiac CT as a running example of how novel machine learning (ML)-based radiomic approaches can improve clinical care. The integration of ML, deep learning, and radiomic methods has revealed direct links between tissue imaging phenotyping and tissue biology, with important clinical implications. More specifically, we discuss the current evidence, strengths, limitations, and future directions for AI in cardiac imaging and CT, as well as lessons that can be learned from other areas. Finally, we propose a scientific framework in order to ensure the clinical and scientific validity of future studies in this novel, yet highly promising field. Still in its infancy, AI-based cardiovascular imaging has a lot to offer to both the patients and their doctors as it catalyzes the transition towards a more precise phenotyping of cardiovascular disease.
    Keywords:  Artificial intelligence; Atherosclerosis; Computed tomography; Plaque; Radiomics; Risk prediction
    DOI:  https://doi.org/10.1093/cvr/cvaa021
  2. Circ Heart Fail. 2020 Mar;13(3): e006513
    Stehlik J, Schmalfuss C, Bozkurt B, Nativi-Nicolau J, Wohlfahrt P, Wegerich S, Rose K, Ray R, Schofield R, Deswal A, Sekaric J, Anand S, Richards D, Hanson H, Pipke M, Pham M.
      BACKGROUND: Implantable cardiac sensors have shown promise in reducing rehospitalization for heart failure (HF), but the efficacy of noninvasive approaches has not been determined. The objective of this study was to determine the accuracy of noninvasive remote monitoring in predicting HF rehospitalization.METHODS: The LINK-HF study (Multisensor Non-invasive Remote Monitoring for Prediction of Heart Failure Exacerbation) examined the performance of a personalized analytical platform using continuous data streams to predict rehospitalization after HF admission. Study subjects were monitored for up to 3 months using a disposable multisensor patch placed on the chest that recorded physiological data. Data were uploaded continuously via smartphone to a cloud analytics platform. Machine learning was used to design a prognostic algorithm to detect HF exacerbation. Clinical events were formally adjudicated.
    RESULTS: One hundred subjects aged 68.4±10.2 years (98% male) were enrolled. After discharge, the analytical platform derived a personalized baseline model of expected physiological values. Differences between baseline model estimated vital signs and actual monitored values were used to trigger a clinical alert. There were 35 unplanned nontrauma hospitalization events, including 24 worsening HF events. The platform was able to detect precursors of hospitalization for HF exacerbation with 76% to 88% sensitivity and 85% specificity. Median time between initial alert and readmission was 6.5 (4.2-13.7) days.
    CONCLUSIONS: Multivariate physiological telemetry from a wearable sensor can provide accurate early detection of impending rehospitalization with a predictive accuracy comparable to implanted devices. The clinical efficacy and generalizability of this low-cost noninvasive approach to rehospitalization mitigation should be further tested. Registration: URL: https://www.clinicaltrials.gov. Unique Identifier: NCT03037710.
    Keywords:  heart failure; hospitalization; machine learning; smartphone; telemetry
    DOI:  https://doi.org/10.1161/CIRCHEARTFAILURE.119.006513
  3. J Am Heart Assoc. 2020 Mar 03. 9(5): e013958
    Han D, Kolli KK, Al'Aref SJ, Baskaran L, van Rosendael AR, Gransar H, Andreini D, Budoff MJ, Cademartiri F, Chinnaiyan K, Choi JH, Conte E, Marques H, de Araújo Gonçalves P, Gottlieb I, Hadamitzky M, Leipsic JA, Maffei E, Pontone G, Raff GL, Shin S, Kim YJ, Lee BK, Chun EJ, Sung JM, Lee SE, Virmani R, Samady H, Stone P, Narula J, Berman DS, Bax JJ, Shaw LJ, Lin FY, Min JK, Chang HJ.
      Background Rapid coronary plaque progression (RPP) is associated with incident cardiovascular events. To date, no method exists for the identification of individuals at risk of RPP at a single point in time. This study integrated coronary computed tomography angiography-determined qualitative and quantitative plaque features within a machine learning (ML) framework to determine its performance for predicting RPP. Methods and Results Qualitative and quantitative coronary computed tomography angiography plaque characterization was performed in 1083 patients who underwent serial coronary computed tomography angiography from the PARADIGM (Progression of Atherosclerotic Plaque Determined by Computed Tomographic Angiography Imaging) registry. RPP was defined as an annual progression of percentage atheroma volume ≥1.0%. We employed the following ML models: model 1, clinical variables; model 2, model 1 plus qualitative plaque features; model 3, model 2 plus quantitative plaque features. ML models were compared with the atherosclerotic cardiovascular disease risk score, Duke coronary artery disease score, and a logistic regression statistical model. 224 patients (21%) were identified as RPP. Feature selection in ML identifies that quantitative computed tomography variables were higher-ranking features, followed by qualitative computed tomography variables and clinical/laboratory variables. ML model 3 exhibited the highest discriminatory performance to identify individuals who would experience RPP when compared with atherosclerotic cardiovascular disease risk score, the other ML models, and the statistical model (area under the receiver operating characteristic curve in ML model 3, 0.83 [95% CI 0.78-0.89], versus atherosclerotic cardiovascular disease risk score, 0.60 [0.52-0.67]; Duke coronary artery disease score, 0.74 [0.68-0.79]; ML model 1, 0.62 [0.55-0.69]; ML model 2, 0.73 [0.67-0.80]; all P<0.001; statistical model, 0.81 [0.75-0.87], P=0.128). Conclusions Based on a ML framework, quantitative atherosclerosis characterization has been shown to be the most important feature when compared with clinical, laboratory, and qualitative measures in identifying patients at risk of RPP.
    Keywords:  coronary artery disease; coronary computed tomography angiography; machine learning; plaque progression; risk prediction
    DOI:  https://doi.org/10.1161/JAHA.119.013958
  4. J Clin Med. 2020 Feb 19. pii: E572. [Epub ahead of print]9(2):
    Díez-Sanmartín C, Sarasa Cabezuelo A.
      A key issue in the field of kidney transplants is the analysis of transplant recipients' survival. By means of the information obtained from transplant patients, it is possible to analyse in which cases a transplant has a higher likelihood of success and the factors on which it will depend. In general, these analyses have been conducted by applying traditional statistical techniques, as the amount and variety of data available about kidney transplant processes were limited. However, two main changes have taken place in this field in the last decade. Firstly, the digitalisation of medical information through the use of electronic health records (EHRs), which store patients' medical histories electronically. This facilitates automatic information processing through specialised software. Secondly, medical Big Data has provided access to vast amounts of data on medical processes. The information currently available on kidney transplants is huge and varied by comparison to that initially available for this kind of study. This new context has led to the use of other non-traditional techniques more suitable to conduct survival analyses in these new conditions. Specifically, this paper provides a review of the main machine learning methods and tools that are being used to conduct kidney transplant patient and graft survival analyses.
    Keywords:  artificial intelligence; kidney transplantation; machine learning; survival
    DOI:  https://doi.org/10.3390/jcm9020572
  5. Acad Radiol. 2020 Feb 24. pii: S1076-6332(20)30084-2. [Epub ahead of print]
    Rao B, Zohrabian V, Cedeno P, Saha A, Pahade J, Davis MA.
      RATIONALE AND OBJECTIVES: Misdiagnosis of intracranial hemorrhage (ICH) can adversely impact patient outcomes. The increasing workload on the radiologists may increase the chance of error and compromise the quality of care provided by the radiologists.MATERIALS AND METHODS: We used an FDA approved artificial intelligence (AI) solution based on a convolutional neural network to assess the prevalence of ICH in scans, which were reported as negative for ICH. We retrospectively applied the AI solution to all consecutive noncontrast computed tomography (CT) head scans performed at eight imaging sites affiliated to our institution.
    RESULTS: In the 6565 noncontrast CT head scans, which met the inclusion criteria, 5585 scans were reported to have no ICH ("negative-by-report" cases). We applied AI solution to these "negative-by-report" cases. AI solution suggested there were ICH in 28 of these scans ("negative-by-report" and "positive-by-AI solution"). After consensus review by three neuroradiologists, 16 of these scans were found to have ICH, which was not reported (missed diagnosis by radiologists), with a false-negative rate of radiologists for ICH detection at 1.6%. Most commonly missed ICH was overlying the cerebral convexity and in the parafalcine regions.
    CONCLUSION: Our study demonstrates that an AI solution can help radiologists to diagnose ICH and thus decrease the error rate. AI solution can serve as a prospective peer review tool for non-contrast head CT scans to identify ICH and thus minimize false negatives.
    Keywords:  Artificial intelligence; Intracranial hemorrhage; Peer-review
    DOI:  https://doi.org/10.1016/j.acra.2020.01.035
  6. Clin Ophthalmol. 2020 ;14 419-429
    Zapata MA, Royo-Fibla D, Font O, Vela JI, Marcantonio I, Moya-Sánchez EU, Sánchez-Pérez A, Garcia-Gasulla D, Cortés U, Ayguadé E, Labarta J.
      Purpose: To assess the performance of deep learning algorithms for different tasks in retinal fundus images: (1) detection of retinal fundus images versus optical coherence tomography (OCT) or other images, (2) evaluation of good quality retinal fundus images, (3) distinction between right eye (OD) and left eye (OS) retinal fundus images,(4) detection of age-related macular degeneration (AMD) and (5) detection of referable glaucomatous optic neuropathy (GON).Patients and Methods: Five algorithms were designed. Retrospective study from a database of 306,302 images, Optretina's tagged dataset. Three different ophthalmologists, all retinal specialists, classified all images. The dataset was split per patient in a training (80%) and testing (20%) splits. Three different CNN architectures were employed, two of which were custom designed to minimize the number of parameters with minimal impact on its accuracy. Main outcome measure was area under the curve (AUC) with accuracy, sensitivity and specificity.
    Results: Determination of retinal fundus image had AUC of 0.979 with an accuracy of 96% (sensitivity 97.7%, specificity 92.4%). Determination of good quality retinal fundus image had AUC of 0.947, accuracy 91.8% (sensitivity 96.9%, specificity 81.8%). Algorithm for OD/OS had AUC 0.989, accuracy 97.4%. AMD had AUC of 0.936, accuracy 86.3% (sensitivity 90.2% specificity 82.5%), GON had AUC of 0.863, accuracy 80.2% (sensitivity 76.8%, specificity 83.8%).
    Conclusion: Deep learning algorithms can differentiate a retinal fundus image from other images. Algorithms can evaluate the quality of an image, discriminate between right or left eye and detect the presence of AMD and GON with a high level of accuracy, sensitivity and specificity.
    Keywords:  artificial intelligence; retinal diseases; retinal fundus image; screening
    DOI:  https://doi.org/10.2147/OPTH.S235751
  7. J Vasc Surg. 2020 Feb 21. pii: S0741-5214(19)32911-8. [Epub ahead of print]
    Raffort J, Adam C, Carrier M, Ballaith A, Coscas R, Jean-Baptiste E, Hassen-Khodja R, Chakfé N, Lareyre F.
      OBJECTIVE: Abdominal aortic aneurysm (AAA) is a life-threatening disease, and the only curative treatment relies on open or endovascular repair. The decision to treat relies on the evaluation of the risk of AAA growth and rupture, which can be difficult to assess in practice. Artificial intelligence (AI) has revealed new insights into the management of cardiovascular diseases, but its application in AAA has so far been poorly described. The aim of this review was to summarize the current knowledge on the potential applications of AI in patients with AAA.METHODS: A comprehensive literature review was performed. The MEDLINE database was searched according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines. The search strategy used a combination of keywords and included studies using AI in patients with AAA published between May 2019 and January 2000. Two authors independently screened titles and abstracts and performed data extraction. The search of published literature identified 34 studies with distinct methodologies, aims, and study designs.
    RESULTS: AI was used in patients with AAA to improve image segmentation and for quantitative analysis and characterization of AAA morphology, geometry, and fluid dynamics. AI allowed computation of large data sets to identify patterns that may be predictive of AAA growth and rupture. Several predictive and prognostic programs were also developed to assess patients' postoperative outcomes, including mortality and complications after endovascular aneurysm repair.
    CONCLUSIONS: AI represents a useful tool in the interpretation and analysis of AAA imaging by enabling automatic quantitative measurements and morphologic characterization. It could be used to help surgeons in preoperative planning. AI-driven data management may lead to the development of computational programs for the prediction of AAA evolution and risk of rupture as well as postoperative outcomes. AI could also be used to better evaluate the indications and types of surgical treatment and to plan the postoperative follow-up. AI represents an attractive tool for decision-making and may facilitate development of personalized therapeutic approaches for patients with AAA.
    Keywords:  Abdominal aortic aneurysm; Aneurysm; Artificial intelligence; Deep learning; EVAR; Endovascular aneurysm repair; Machine learning; Open repair
    DOI:  https://doi.org/10.1016/j.jvs.2019.12.026
  8. Res Pract Thromb Haemost. 2020 Feb;4(2): 230-237
    Nafee T, Gibson CM, Travis R, Yee MK, Kerneis M, Chi G, AlKhalfan F, Hernandez AF, Hull RD, Cohen AT, Harrington RA, Goldhaber SZ.
      Background: The identification of acutely ill patients at high risk for venous thromboembolism (VTE) may be determined clinically or by use of integer-based scoring systems. These scores demonstrated modest performance in external data sets.Objectives: To evaluate the performance of machine learning models compared to the IMPROVE score.
    Methods: The APEX trial randomized 7513 acutely medically ill patients to extended duration betrixaban vs. enoxaparin. Including 68 variables, a super learner model (ML) was built to predict VTE by combining estimates from 5 families of candidate models. A "reduced" model (rML) was also developed using 16 variables that were thought, a priori, to be associated with VTE. The IMPROVE score was calculated for each patient. Model performance was assessed by discrimination and calibration to predict a composite VTE end point. The frequency of predicted risks of VTE were plotted and divided into tertiles. VTE risks were compared across tertiles.
    Results: The ML and rML algorithms outperformed the IMPROVE score in predicting VTE (c-statistic: 0.69, 0.68 and 0.59, respectively). The Hosmer-Lemeshow goodness-of-fit P-value was 0.06 for ML, 0.44 for rML, and <0.001 for the IMPROVE score. The observed event rate in the lowest tertile was 2.5%, 4.8% in tertile 2, and 11.4% in the highest tertile. Patients in the highest tertile of VTE risk had a 5-fold increase in odds of VTE compared to the lowest tertile.
    Conclusion: The super learner algorithms improved discrimination and calibration compared to the IMPROVE score for predicting VTE in acute medically ill patients.
    Keywords:  acute medically ill; machine learning; personalized medicine; super learner; venous thromboembolism
    DOI:  https://doi.org/10.1002/rth2.12292
  9. J Invest Dermatol. 2020 Mar;pii: S0022-202X(20)30063-4. [Epub ahead of print]140(3): 507-514.e1
    Cullell-Dalmau M, Otero-Viñas M, Manzo C.
      Deep learning is a branch of artificial intelligence that uses computational networks inspired by the human brain to extract patterns from raw data. Development and application of deep learning methods for image analysis, including classification, segmentation, and restoration, have accelerated in the last decade. These tools have been progressively incorporated into several research fields, opening new avenues in the analysis of biomedical imaging. Recently, the application of deep learning to dermatological images has shown great potential. Deep learning algorithms have shown performance comparable with humans in classifying skin lesion images into different skin cancer categories. The potential relevance of deep learning to the clinical realm created the need for researchers in disciplines other than computer science to understand its fundamentals. In this paper, we introduce the basics of a deep learning architecture for image classification, the convolutional neural network, in a manner accessible to nonexperts. We explain its fundamental operation, the convolution, and describe the metrics for the evaluation of its performance. These concepts are important to interpret and evaluate scientific publications involving these tools. We also present examples of recent applications for dermatology. We further discuss the capabilities and limitations of these artificial intelligence-based methods.
    DOI:  https://doi.org/10.1016/j.jid.2019.12.029
  10. Lancet Respir Med. 2020 Feb 25. pii: S2213-2600(20)30003-5. [Epub ahead of print]
    Walsh SLF, Humphries SM, Wells AU, Brown KK.
      Over the past decade, there has been a groundswell of research interest in computer-based methods for objectively quantifying fibrotic lung disease on high resolution CT of the chest. In the past 5 years, the arrival of deep learning-based image analysis has created exciting new opportunities for enhancing the understanding of, and the ability to interpret, fibrotic lung disease on CT. Specific unsolved problems for which computer-based imaging analysis might provide solutions include the development of reliable methods for assisting with diagnosis, detecting early disease, and predicting disease behaviour using baseline imaging data. However, to harness this technology, technical and societal challenges must be overcome. Large CT datasets will be needed to power the training of deep learning algorithms. Open science research and collaboration between academia and industry must be encouraged. Prospective clinical utility studies will be needed to test computer algorithm performance in real-world clinical settings and demonstrate patient benefit over current best practice. Finally, ethical standards, which ensure patient confidentiality and mitigate against biases in training datasets, that can be encoded in machine-learning systems will be needed as well as bespoke data governance and accountability frameworks to encourage buy-in from health-care professionals, patients, and the public.
    DOI:  https://doi.org/10.1016/S2213-2600(20)30003-5
  11. PLoS One. 2020 ;15(2): e0229596
    Mirchi N, Bissonnette V, Yilmaz R, Ledwos N, Winkler-Schwartz A, Del Maestro RF.
      Simulation-based training is increasingly being used for assessment and training of psychomotor skills involved in medicine. The application of artificial intelligence and machine learning technologies has provided new methodologies to utilize large amounts of data for educational purposes. A significant criticism of the use of artificial intelligence in education has been a lack of transparency in the algorithms' decision-making processes. This study aims to 1) introduce a new framework using explainable artificial intelligence for simulation-based training in surgery, and 2) validate the framework by creating the Virtual Operative Assistant, an automated educational feedback platform. Twenty-eight skilled participants (14 staff neurosurgeons, 4 fellows, 10 PGY 4-6 residents) and 22 novice participants (10 PGY 1-3 residents, 12 medical students) took part in this study. Participants performed a virtual reality subpial brain tumor resection task on the NeuroVR simulator using a simulated ultrasonic aspirator and bipolar. Metrics of performance were developed, and leave-one-out cross validation was employed to train and validate a support vector machine in Matlab. The classifier was combined with a unique educational system to build the Virtual Operative Assistant which provides users with automated feedback on their metric performance with regards to expert proficiency performance benchmarks. The Virtual Operative Assistant successfully classified skilled and novice participants using 4 metrics with an accuracy, specificity and sensitivity of 92, 82 and 100%, respectively. A 2-step feedback system was developed to provide participants with an immediate visual representation of their standing related to expert proficiency performance benchmarks. The educational system outlined establishes a basis for the potential role of integrating artificial intelligence and virtual reality simulation into surgical educational teaching. The potential of linking expertise classification, objective feedback based on proficiency benchmarks, and instructor input creates a novel educational tool by integrating these three components into a formative educational paradigm.
    DOI:  https://doi.org/10.1371/journal.pone.0229596
  12. Comput Methods Programs Biomed. 2020 Jan 16. pii: S0169-2607(19)31101-0. [Epub ahead of print]191 105320
    Islam MM, Yang HC, Poly TN, Jian WS, Jack Li YC.
      BACKGROUND: Diabetic retinopathy (DR) is one of the leading causes of blindness globally. Earlier detection and timely treatment of DR are desirable to reduce the incidence and progression of vision loss. Currently, deep learning (DL) approaches have offered better performance in detecting DR from retinal fundus images. We, therefore, performed a systematic review with a meta-analysis of relevant studies to quantify the performance of DL algorithms for detecting DR.METHODS: A systematic literature search on EMBASE, PubMed, Google Scholar, Scopus was performed between January 1, 2000, and March 31, 2019. The search strategy was based on the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) reporting guidelines, and DL-based study design was mandatory for articles inclusion. Two independent authors screened abstracts and titles against inclusion and exclusion criteria. Data were extracted by two authors independently using a standard form and the Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) tool was used for the risk of bias and applicability assessment.
    RESULTS: Twenty-three studies were included in the systematic review; 20 studies met inclusion criteria for the meta-analysis. The pooled area under the receiving operating curve (AUROC) of DR was 0.97 (95%CI: 0.95-0.98), sensitivity was 0.83 (95%CI: 0.83-0.83), and specificity was 0.92 (95%CI: 0.92-0.92). The positive- and negative-likelihood ratio were 14.11 (95%CI: 9.91-20.07), and 0.10 (95%CI: 0.07-0.16), respectively. Moreover, the diagnostic odds ratio for DL models was 136.83 (95%CI: 79.03-236.93). All the studies provided a DR-grading scale, a human grader (e.g. trained caregivers, ophthalmologists) as a reference standard.
    CONCLUSION: The findings of our study showed that DL algorithms had high sensitivity and specificity for detecting referable DR from retinal fundus photographs. Applying a DL-based automated tool of assessing DR from color fundus images could provide an alternative solution to reduce misdiagnosis and improve workflow. A DL-based automated tool offers substantial benefits to reduce screening costs, accessibility to healthcare and ameliorate earlier treatments.
    Keywords:  Deep learning; Diabetic; Diabetic retinopathy; Fundus photograph; Retinopathy
    DOI:  https://doi.org/10.1016/j.cmpb.2020.105320
  13. Med Phys. 2020 Feb 25.
    Kadoya N, Tanaka S, Kajikawa T, Tanabe S, Abe K, Nakajima Y, Yamamoto T, Takahashi N, Takeda K, Dobashi S, Takeda K, Nakane K, Jingu K.
      PURPOSE: Radiomics is a new technique that enables noninvasive prognostic prediction by extracting features from medical images. Homology is a concept used in many branches of algebra and topology that can quantify the contact degree. In the present study, we developed homology-based radiomic features to predict the prognosis of non-small-cell lung cancer (NSCLC) patients and then evaluated the accuracy of this prediction method.METHODS: Four data sets were used: two to provide training and test data and two for the selection of robust radiomic features. All the data sets were downloaded from The Cancer Imaging Archive (TCIA). In two-dimensional cases, the Betti numbers consist of two values: b0 (zero-dimensional Betti number), which is the number of isolated components, and b1 (one-dimensional Betti number), which is the number of one-dimensional or "circular" holes. For homology-based evaluation, CT images must be converted to binarized images in which each pixel has two possible values: 0 or 1. All CT slices of the gross tumor volume were used for calculating the homology histogram. First, by changing the threshold of the CT value (range: -150 to 300 HU) for all its slices, we developed homology-based histograms for b0 , b1 , and b1 /b0 using binarized images All histograms were then summed, and the summed histogram was normalized by the number of slices. 144 homology-based radiomic features were defined from the histogram. To compare the standard radiomic features, 107 radiomic features were calculated using the standard radiomics technique. To clarify the prognostic power, the relationship between the values of the homology-based radiomic features and overall survival was evaluated using LASSO Cox regression model and the Kaplan-Meier method. The retained features with non-zero coefficients calculated by the LASSO Cox regression model were used for fitting the regression model. Moreover, these features were then integrated into a radiomics signature. An individualized rad score was calculated from a linear combination of the selected features, which were weighted by their respective coefficients.
    RESULTS: When the patients in the training and test data sets were stratified into high-risk and low-risk groups according to the rad scores, the overall survival of the groups was significantly different. The C-index values for the homology-based features (rad score), standard features (rad score), and tumor size were 0.625, 0.603, and 0.607, respectively, for the training data sets and 0.689, 0.668, and 0.667 for the test data sets. This result showed that homology-based radiomic features had slightly higher prediction power than the standard radiomic features.
    CONCLUSIONS: Prediction performance using homology-based radiomic features had a comparable or slightly higher prediction power than standard radiomic features. These findings suggest that homology-based radiomic features may have great potential for improving the prognostic prediction accuracy of CT-based radiomics. In this result, it is noteworthy that there are some limitations.
    Keywords:  CT image; lung cancer; machine learning; radiomics; survival
    DOI:  https://doi.org/10.1002/mp.14104
  14. Hepatology. 2020 Feb 28.
    Saillard C, Schmauch B, Laifa O, Moarii M, Toldo S, Zaslavskiy M, Pronier E, Laurent A, Amaddeo G, Regnault H, Sommacale D, Ziol M, Pawlotsky JM, Mulé S, Luciani A, Wainrib G, Clozel T, Courtiol P, Calderaro J.
      Standardized and robust risk stratification systems for patients with hepatocellular carcinoma (HCC) are required to improve therapeutic strategies and investigate the benefits of adjuvant systemic therapies after curative resection/ablation. In this study, we used two deep-learning algorithms based on whole-slide digitized histological slides (WSI) to build models for predicting the survival of patients with HCC treated by surgical resection. Two independent series were investigated: a discovery set (Henri Mondor Hospital, n=194) used to develop our algorithms and an independent validation set (TCGA, n=328). WSIs were first divided into small squares ("tiles") and features were extracted with a pretrained convolutional neural network (preprocessing step). The first deep-learning based algorithm ("SCHMOWDER") uses an attention mechanism on tumoral areas annotated by a pathologist while the second ("CHOWDER") does not require human expertise. In the discovery set, c-indexes for survival prediction of SCHMOWDER and CHOWDER reached 0.78 and 0.75, respectively. Both models outperformed a composite score incorporating all baseline variables associated with survival. The prognostic value of the models was further validated in the TCGA dataset, and, as observed in the discovery series, both models had a higher discriminatory power than a score combining all baseline variables associated with survival. Pathological review showed that the tumoral areas most predictive of poor survival were characterized by vascular spaces, the macrotrabecular architectural pattern and a lack of immune infiltration. CONCLUSION: This study shows that artificial intelligence can help refine the prediction of HCC prognosis. It highlights the importance of pathologist/machine interactions for the construction of deep-learning algorithms that benefit from expert knowledge and allow a biological understanding of their output.
    Keywords:  artificial intelligence; deep-learning; hepatocellular carcinoma; survival
    DOI:  https://doi.org/10.1002/hep.31207
  15. Acad Radiol. 2020 Feb 20. pii: S1076-6332(19)30626-9. [Epub ahead of print]
    Dontchos BN, Yala A, Barzilay R, Xiang J, Lehman CD.
      RATIONALE AND OBJECTIVES: Federal legislation requires patient notification of dense mammographic breast tissue because increased density is a marker of breast cancer risk and can limit the sensitivity of mammography. As previously described, we clinically implemented our deep learning model at the academic breast imaging practice where the model was developed with high clinical acceptance. Our objective was to externally validate our deep learning model on radiologist breast density assessments in a community breast imaging practice.MATERIALS AND METHODS: Our deep learning model was implemented at a dedicated breast imaging practice staffed by both academic and community breast imaging radiologists in October 2018. Deep learning model assessment of mammographic breast density was presented to the radiologist during routine clinical practice at the time of mammogram interpretation. We identified 2174 consecutive screening mammograms after implementation of the deep learning model. Radiologist agreement with the model's assessment was measured and compared across radiologist groups.
    RESULTS: Both academic and community radiologists had high clinical acceptance of the deep learning model's density prediction, with 94.9% (academic) and 90.7% (community) acceptance for dense versus nondense categories (p < 0.001). The proportion of mammograms assessed as dense by all radiologists decreased from 47.0% before deep learning model implementation to 41.0% after deep learning model implementation (p < 0.001).
    CONCLUSION: Our deep learning model had a high clinical acceptance rate among both academic and community radiologists and reduced the proportion of mammograms assessed as dense. This is an important step to validating our deep learning model prior to potential widespread implementation.
    Keywords:  Breast Density; Deep Learning; Mammography
    DOI:  https://doi.org/10.1016/j.acra.2019.12.012
  16. PLoS One. 2020 ;15(2): e0226157
    Lee S, Liang X, Woods M, Reiner AS, Concannon P, Bernstein L, Lynch CF, Boice JD, Deasy JO, Bernstein JL, Oh JH.
      The purpose of this study was to identify germline single nucleotide polymorphisms (SNPs) that optimally predict radiation-associated contralateral breast cancer (RCBC) and to provide new biological insights into the carcinogenic process. Fifty-two women with contralateral breast cancer and 153 women with unilateral breast cancer were identified within the Women's Environmental Cancer and Radiation Epidemiology (WECARE) Study who were at increased risk of RCBC because they were ≤ 40 years of age at first diagnosis of breast cancer and received a scatter radiation dose > 1 Gy to the contralateral breast. A previously reported algorithm, preconditioned random forest regression, was applied to predict the risk of developing RCBC. The resulting model produced an area under the curve (AUC) of 0.62 (p = 0.04) on hold-out validation data. The biological analysis identified the cyclic AMP-mediated signaling and Ephrin-A as significant biological correlates, which were previously shown to influence cell survival after radiation in an ATM-dependent manner. The key connected genes and proteins that are identified in this analysis were previously identified as relevant to breast cancer, radiation response, or both. In summary, machine learning/bioinformatics methods applied to genome-wide genotyping data have great potential to reveal plausible biological correlates associated with the risk of RCBC.
    DOI:  https://doi.org/10.1371/journal.pone.0226157
  17. N Engl J Med. 2020 Feb 27. 382(9): 791-793
    Mehta MC, Katz IT, Jha AK.
      
    DOI:  https://doi.org/10.1056/NEJMp1912079
  18. Data Brief. 2020 Apr;29 105213
    Lanka P, Rangaprakash D, Gotoor SSR, Dretsch MN, Katz JS, Denney TS, Deshpande G.
      Resting-state functional Magnetic Resonance Imaging (rs-fMRI) has been extensively used for diagnostic classification because it does not require task compliance and is easier to pool data from multiple imaging sites, thereby increasing the sample size. A MATLAB-based toolbox called Machine Learning in NeuroImaging (MALINI) for feature extraction and disease classification is presented. The MALINI toolbox extracts functional and effective connectivity features from preprocessed rs-fMRI data and performs classification between healthy and disease groups using any of 18 popular and widely used machine learning algorithms that are based on diverse principles. A consensus classifier combining the power of multiple classifiers is also presented. The utility of the toolbox is illustrated by accompanying data consisting of resting-state functional connectivity features from healthy controls and subjects with various brain-based disorders: autism spectrum disorder from autism brain imaging data exchange (ABIDE), Alzheimer's disease and mild cognitive impairment from Alzheimer's disease neuroimaging initiative (ADNI), attention deficit hyperactivity disorder from ADHD-200, and post-traumatic stress disorder and post-concussion syndrome acquired in-house. Results of classification performed on the above datasets can be obtained from the main article titled "Supervised machine learning for diagnostic classification from large-scale neuroimaging datasets" [1]. The data was divided into homogeneous and heterogeneous splits, such that 80% could be used for training, model building and cross-validation, while the remaining 20% of the data could be used as a hold-out independent test data for replication of the classification performance, to ensure the robustness of the classifiers to population variance in image acquisition site and age of the sample.
    Keywords:  ADHD; Alzheimer's disease; Autism; Diagnostic classification; Functional connectivity; PTSD; Resting-state functional MRI; Supervised machine learning
    DOI:  https://doi.org/10.1016/j.dib.2020.105213