Clinical Evidence Extraction from Electronic Health Records and - PowerPoint PPT Presentation

Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature Hamed Hassanzadeh and Anthony Nguyen 31 July 2018 THE AUSTRALIAN E-HEALTH RESEARCH CENTRE

Problem • Radiology reports • Discharge summaries • Progress notes • Biomedical publications Looking for • Different tasks relevant • Different clinical evidence information 2 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

AI Solution Publication A Publications and EHRs Related to Patient Related to Publications Publication B Similar to EHRs Publication C 3 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Methodology • Deep Learning: based on Artificial Neural Networks that are biologically-inspired paradigms • Enables a computer to learn from observed data 4 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Methodology (cont.) Why Deep Learning? • Conventional Machine Learning Approaches: – Task-specific feature engineering – Institution-centric design – Transferability issue • Deep Learning: – Numerical representations as input – More generalizable – Transferable 5 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Methodology (cont.) • Word representation – Word2vec • Deep Learning – Convolutional Neural Network 6 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Methodology (cont.) • Convolutional Neural Networks (CNN) Convolutional layer Fully connected with There layer multiple filter widths is a fracture at the distal shaft … Target Classes Vector representations of Max-pooling layer each word in the document (Embedding Layer) 7 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Data • Four datasets: DATA SET TYPE OF EVIDENCE TYPE OF DOCUMENTS • Medical Test i2b2-2010 Clinical documents (progress reports) • Problem • Treatment • Disorder ShARe/CLEF Clinical documents (discharge summaries) • Intervention NICTA-PIBOSO Biomedical publications (abstracts) • Problem/Population • Outcome • Abnormality ED-Radiology Clinical Document (Radiology reports – three different hospitals) • Benchmark • Support Vector Machines (SVM) • Random Forest (RF) 8 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Results • Results on public data – Goal: validate generalisability (multiple tasks) I2B2-2010 SHARE/CLEF NICTA-PIBOSO TREATMENT PROBLEM TEST DISORDER INTERVENTION POPULATION OUTCOME 0.59 0.65 0.73 0.74 0.02 0.0 0.51 RF 0.80 0.84 0.85 0.85 0.0 0.21 0.68 SVM 0.90 0.93 0.93 0.87 0.41 0.57 0.75 Our Approach 9 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Results (cont.) • Results on ED-Radiology – Goal: validate transferability (multiple institutions) RBWH RCH GCH RF 0.80 0.83 0.84 SVM 0.83 0.91 0.91 Our approach 0.91 0.94 0.94 10 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Conclusion • A generalizable and transferable solution – Different clinical document (progress notes, discharge summaries) – Biomedical publications – Different institutions • Future work: – Expand the solution and validate it over more and bigger clinical problems – Translating the outcome more into practice 11 | Clinical Evidence Extraction from Electronic Health Records and Biomedical Literature| Hamed Hassanzadeh

Thank you Hamed Hassanzadeh, Postdoctoral Fellow e hamed.hassanzadeh@csiro.au HEALTH & BIOSECURITY