Build the Next Cookbook
We're looking for ML Engineers and Technical Writers to help us expand this directory of open-source medical AI recipes.
Open AI Playground →Cookbook 1: The Clinical Reasoning Agent🔗
Deploy an intelligent clinical NLP bot capable of parsing and diagnostic reasoning against structured FHIR patient records entirely locally.
Synthea
Simulated synthetic patient populations exported in pure JSON FHIR bundles.
Medplum
An open-source FHIR server to ingest and serve the Synthea data via a standard REST API.
BioMistral
A locally quantized open-weights LLM fine-tuned specifically on medical literature.
Expected Outcome:
You will have a local React application querying your Medplum FHIR server, taking patient symptoms, and feeding them to BioMistral running in `llama.cpp` to output diagnostic differentials.
$ openphr deploy the-clinical-reasoning-agent
Cookbook 2: Local Medical Imaging Pipeline🔗
Train a customized Convolutional Neural Network (CNN) to detect anomalies in X-rays or MRI scans natively on your GPU infrastructure.
Stanford CheXpert
A large public dataset of chest X-rays with structured anomaly labels.
MONAI
A PyTorch-based framework optimized specifically for reading DICOM files and deep learning in healthcare imaging.
Expected Outcome:
A deployable Docker container utilizing MONAI to intake raw patient DICOM scans and output a semantic segmentation map highlighting potential pneumonia clusters.
$ openphr deploy local-medical-imaging-pipeline
Cookbook 3: Real-Time Wearable Anomaly Detection🔗
Process live ECG and heart rate data from wearables to detect arrhythmias using lightweight edge AI models.
MIT-BIH Arrhythmia Database
Standard test material for evaluating arrhythmia detectors, available via PhysioNet.
TensorFlow Lite Micro
A highly compressed 1D-CNN designed to run on microcontrollers with milliwatt power consumption.
Expected Outcome:
A deployable edge model on a Raspberry Pi or smartwatch identifying premature ventricular contractions in real time.
$ openphr deploy real-time-wearable-anomaly-detection
Cookbook 4: Genomic Variant Classification🔗
Build a pipeline to predict the clinical pathogenicity of newly discovered genetic mutations.
ClinVar
Public archive of reports of the relationships among human variations and phenotypes.
Hail
Open-source, scalable framework for exploring and analyzing genomic data at scale.
AlphaMissense
DeepMind's model for predicting the effect of missense variants.
Expected Outcome:
A scalable Spark cluster using Hail to filter patient VCFs and scoring variants with AlphaMissense to flag rare disease candidates.
$ openphr deploy genomic-variant-classification
Cookbook 5: Oncology Pathology Slide Segmentation🔗
Automate the counting and classification of cancer cells in gigapixel whole slide images (WSIs).
Camelyon16
Breast cancer metastasis detection dataset in sentinel lymph nodes.
OpenSlide
A C library that provides a simple interface to read whole-slide images.
Expected Outcome:
A tile-based pipeline extracting high-res patches from WSIs, processing them via a vision transformer (ViT), and generating a tumor-probability heatmap overlay.
$ openphr deploy oncology-pathology-slide-segmentation
Cookbook 6: Medical Audio Dictation Engine🔗
Build a private, HIPAA-compliant voice-to-text service that automatically transcribes doctor-patient encounters.
Whisper (Medical Fine-Tune)
OpenAI's robust ASR model fine-tuned on complex medical jargon and pharmaceutical names.
CTranslate2
A fast inference engine for Transformer models, executing Whisper locally in real-time.
Expected Outcome:
A local microphone streaming service transcribing clinical encounters directly into the EMR without sending audio to the cloud.
$ openphr deploy medical-audio-dictation-engine
Cookbook 7: Predicting Patient Readmission🔗
Train tabular machine learning models on longitudinal EMR data to flag high-risk discharges.
MIMIC-IV
A freely accessible clinical database from the Beth Israel Deaconess Medical Center ICU.
XGBoost
An optimized distributed gradient boosting library highly effective for sparse clinical tabular data.
Expected Outcome:
A predictive scoring system identifying which ICU patients have a >30% probability of 30-day readmission.
$ openphr deploy predicting-patient-readmission
Cookbook 8: Automated Drug Repurposing Discovery🔗
Use knowledge graphs and link prediction to discover new therapeutic uses for existing FDA-approved drugs.
DrugBank / KEGG
Comprehensive databases containing information on drugs, targets, and pathways.
Neo4j
A powerful graph database engine to map complex biological interactions.
Graph Neural Networks (GNNs)
Models that perform link prediction (e.g., Drug A connects to Disease B).
Expected Outcome:
A queryable interface suggesting statistically probable off-label uses for existing hypertension drugs against novel viral variants.
$ openphr deploy automated-drug-repurposing-discovery
Cookbook 9: Robotic Surgery Video Analysis🔗
Track surgical instruments in real-time endoscopic video to evaluate surgeon performance and prevent errors.
EndoVis (Endoscopic Vision Challenge)
A collection of annotated robotic surgery video frames identifying tools and tissue.
YOLOv8
A state-of-the-art, ultra-fast object detection model capable of 60+ FPS inference.
Expected Outcome:
A real-time dashboard overlaying bounding boxes on a live laparoscopy feed, tracking the scalpel and forceps with millimeter precision.
$ openphr deploy robotic-surgery-video-analysis
Cookbook 10: 3D Protein Structure Folding🔗
Predict the 3D atomic structure of a protein from its 1D amino acid sequence to design synthetic antibodies.
PDB (Protein Data Bank)
The global archive of experimentally determined 3D structures of biological macromolecules.
OpenFold
A trainable, open-source reproduction of AlphaFold2 for protein folding inference and fine-tuning.
Expected Outcome:
An automated pipeline outputting .pdb files of folded proteins, viewable in PyMOL, to analyze binding pockets for drug discovery.
$ openphr deploy 3d-protein-structure-folding
Cookbook 11: Longitudinal Cohort Extraction🔗
Transform unstructured clinical notes into structured OMOP common data models for population health research.
i2b2 NLP Datasets
De-identified clinical notes annotated with named entities, temporal relations, and coreferences.
Spark NLP for Healthcare
John Snow Labs' commercial-grade NLP library optimized for clinical entity recognition.
OMOP CDM
The Observational Medical Outcomes Partnership Common Data Model schema.
Expected Outcome:
A distributed Spark job that reads thousands of free-text discharge summaries, extracts ICD-10 and SNOMED codes, and inserts them into a standardized SQL data warehouse.
$ openphr deploy longitudinal-cohort-extraction
Cookbook 12: Real-World Evidence Generation from Social Media🔗
Mine public Reddit and Twitter streams to detect unreported adverse drug reactions in near real-time.
SMM4H
Social Media Mining for Health Applications annotated tweets for adverse drug events.
HuggingFace Transformers
The industry-standard library for downloading and training language models.
ClinicalBERT
A BERT model specifically pre-trained on clinical text.
Expected Outcome:
A pipeline querying the Reddit API, parsing colloquial patient descriptions with ClinicalBERT, and alerting pharmacovigilance teams to novel side effects.
$ openphr deploy real-world-evidence-generation-from-social-media
Cookbook 13: Federated Learning for Multi-Hospital Privacy🔗
Train a global pneumonia detection model across multiple hospitals without ever moving patient data across hospital firewalls.
NVIDIA FLARE
A robust, open-source SDK for building federated learning paradigms.
ResNet-50
A deep convolutional network suitable for sharing gradients instead of raw X-rays.
Expected Outcome:
A central aggregation server that coordinates model weights from decentralized edge-nodes, achieving state-of-the-art accuracy while maintaining total HIPAA compliance.
$ openphr deploy federated-learning-for-multi-hospital-privacy
Cookbook 14: Reinforcement Learning for Personalized Dosing🔗
Use offline reinforcement learning to determine the optimal dosing strategy for sepsis patients in the ICU.
eICU Collaborative Research Database
A multi-center database comprising over 200,000 ICU admissions.
Deep Q-Network (DQN)
An RL algorithm that learns the value of vasopressor dosages given a patient's vital states.
Expected Outcome:
An AI-driven clinician assist tool that recommends real-time fluid and vasopressor adjustments to stabilize blood pressure in septic shock.
$ openphr deploy reinforcement-learning-for-personalized-dosing
Cookbook 15: Automating ICD-10 Coding with LLMs🔗
Eliminate manual medical billing by using Large Language Models to automatically extract correct ICD-10 and CPT codes from discharge summaries.
Clinical LLaMA 3
A quantized LLM tuned specifically for structured information extraction from clinical notes.
LangChain
A framework for developing applications powered by language models with precise JSON-output parsers.
Expected Outcome:
A deployable API that ingests a raw physician note and reliably outputs a strictly formatted JSON array of valid, billable ICD-10 codes.
$ openphr deploy automating-icd-10-coding-with-llms
Cookbook 16: EEG Seizure Prediction🔗
Train a temporal neural network to predict the onset of epileptic seizures minutes before they occur.
CHB-MIT Scalp EEG
Continuous EEG recordings from pediatric subjects with intractable seizures.
LSTM Network
Long Short-Term Memory models excel at finding patterns in sequential time-series data like brainwaves.
Expected Outcome:
A predictive model that processes multi-channel EEG signals and fires an alert 5 minutes prior to clinical seizure onset, allowing for preventative intervention.
$ openphr deploy eeg-seizure-prediction
Cookbook 17: Synthetic Medical Data Generation🔗
Bypass privacy restrictions by using Generative Adversarial Networks (GANs) to create realistic but completely fake patient records.
SDV (Synthetic Data Vault)
An ecosystem of tools to model and generate synthetic tabular datasets.
CTGAN
A GAN architecture specifically designed to handle mixed continuous and discrete tabular variables.
Expected Outcome:
A script that ingests a highly secure proprietary database and outputs millions of rows of safe, shareable synthetic patient data with matching statistical properties.
$ openphr deploy synthetic-medical-data-generation
Cookbook 18: Digital Pathology Survival Prediction🔗
Predict patient overall survival directly from H&E stained pathology slides using multiple-instance learning.
TCGA (The Cancer Genome Atlas)
Vast repository of genomic and histological data matched with patient survival timelines.
CLAM
Clustering-constrained Attention Multiple Instance Learning for gigapixel image classification without localized annotations.
Expected Outcome:
A biomarker discovery tool that identifies highly prognostic morphological regions within tumors and estimates patient risk stratification.
$ openphr deploy digital-pathology-survival-prediction
Cookbook 19: Real-time Bed Capacity Prediction🔗
Forecast hospital census and ICU bed shortages 48 hours in advance to optimize staff scheduling.
Apache Airflow
An open-source platform to programmatically author, schedule, and monitor data pipelines.
Prophet
Meta's robust time-series forecasting algorithm that handles missing data and large outliers well.
Expected Outcome:
An automated morning dashboard for hospital administration highlighting predicted capacity crunches and recommending proactive discharge planning.
$ openphr deploy real-time-bed-capacity-prediction
Cookbook 20: Antimicrobial Resistance (AMR) Prediction🔗
Predict whether a bacterial infection will resist specific antibiotics by sequencing the pathogen genome.
PATRIC
The Pathosystems Resource Integration Center, housing thousands of bacterial genomes and AMR metadata.
Random Forest
An ensemble learning method highly effective at interpreting raw k-mer frequency counts from genomic sequences.
Expected Outcome:
A bioinformatics tool that analyzes rapid genome sequencing from a blood culture and recommends the most effective antibiotic class within hours.
$ openphr deploy antimicrobial-resistance-amr-prediction
Cookbook 21: Drug-Drug Interaction Graph Mining🔗
Identify dangerous polypharmacy side effects by traversing a massive pharmacological knowledge graph.
TWOSIDES
A database of polypharmacy side effects extracted from FDA adverse event reports.
PyTorch Geometric
A geometric deep learning extension library for PyTorch.
Expected Outcome:
An API endpoint for EHR systems that flags severe combination contraindications when a physician attempts to prescribe a new medication.
$ openphr deploy drug-drug-interaction-graph-mining
Cookbook 22: Brain Tumor Segmentation in 3D MRI🔗
Automatically delineate core tumor regions in multi-modal 3D MRI scans to assist neurosurgical planning.
BraTS (Brain Tumor Segmentation)
A premier dataset of MRI scans with expert-annotated tumor sub-regions.
3D U-Net
The canonical architecture for volumetric biomedical image segmentation.
Expected Outcome:
A precise 3D voxel mask that isolates edema, enhancing tumor core, and necrotic regions, exportable as an STL file for 3D printing or AR visualization.
$ openphr deploy brain-tumor-segmentation-in-3d-mri
Build the Future of Open Healthcare
Enjoying these cookbooks? The OpenPHR ecosystem is built entirely by open-source volunteers. We are actively recruiting ML engineers, web developers, and clinical reviewers to help expand these guides.
Try the AI Playground TodayCookbook 23: Contactless Heart Rate Monitoring🔗
Deploy a privacy-first web application that extracts real-time physiological signals (pulse and heart rate) directly from a user's webcam feed using remote photoplethysmography (rPPG).
PulseVision
Our open-source computer vision pipeline that isolates the forehead region and amplifies micro-color changes in human skin.
MediaPipe Face Mesh
A lightweight, on-device ML model that tracks 468 3D facial landmarks in real-time, even on mobile devices.
Expected Outcome:
A client-side web application running entirely in the browser that measures resting heart rate within seconds without sending video data to any server.
$ openphr deploy contactless-heart-rate-monitoring
Cookbook 24: CareHub Clinical Trial Screening Bot🔗
Deploy an intelligent patient-facing chatbot that ingests complex clinical trial eligibility matrices and automatically prescreens patients based on their medical history.
LlamaIndex
A data framework for connecting custom data sources to large language models, perfect for querying complex trial protocols.
CareHub Trial Database
Structured OpenPHR clinical trial matrices for diseases like Parkinson's and Atopic Dermatitis.
Expected Outcome:
A deployable, HIPAA-compliant conversational agent that matches patients to active clinical trials using CareHub data matrices.
$ openphr deploy carehub-clinical-trial-screening-bot
Cookbook 25: Clinical Entity Extraction with BioBERT🔗
Extract critical medical entities (diseases, drugs, dosages, and diagnostic procedures) from unstructured clinical text using a fine-tuned BioBERT model, mapping them directly to standardized SNOMED-CT terminologies.
BioBERT
A domain-specific language representation model pre-trained on large-scale biomedical corpora (PubMed abstracts and PMC full-text articles).
spaCy MedSpaCy Pipeline
A modular clinical NLP library designed to extract concepts and contextual indicators from clinical narratives.
Expected Outcome:
A standardized pipeline that parses unstructured physician notes and returns a structured JSON mapping of symptoms, diagnoses, and treatments.
$ openphr deploy clinical-entity-extraction-biobert
Cookbook 26: HIPAA De-identification of Clinical Narratives🔗
Automatically identify, redact, and replace Protected Health Information (PHI) like patient names, phone numbers, facility addresses, and clinical IDs within narrative electronic health records to ensure HIPAA Safe Harbor compliance.
Clinical-DeID-Llama-3
A fine-tuned Llama model optimized specifically for detecting sensitive PHI entities in medical narratives.
Microsoft Presidio Analyzer
An open-source data protection and de-identification SDK that utilizes custom recognizers and pattern matching.
Expected Outcome:
A secure, local Python script that takes raw clinical summaries, scrubs all 18 HIPAA defined identifiers, and saves safe, shareable research files.
$ openphr deploy patient-phi-deidentification
Cookbook 27: DICOM De-identification and Defacing🔗
Remove sensitive metadata tags from DICOM file headers and deface 3D structural brain MRI images (removing facial contours) to prevent reconstruction of the subject's face while retaining neurological features for machine learning research.
Defaced-Brain-MRI-GAN
A generative adversarial model trained to identify and deface facial features in 3D brain scans without impacting structural brain tissues.
PyDicom & PyDeface
Libraries for reading and modifying DICOM headers and executing coordinate-based defacing algorithms on structural T1/T2 MRI images.
Expected Outcome:
A standardized pre-processing pipeline that outputs fully anonymized and face-scrubbed 3D medical images ready for public model training.
$ openphr deploy dicom-deidentification-defacing
Cookbook 28: Mapping Legacy SQL Data to HL7 FHIR Resources🔗
Transform unstructured or relational SQL/CSV patient records into HL7 FHIR (Fast Healthcare Interoperability Resources) compliant JSON structures, and perform schema validation using local HL7 validators.
FHIR-Mapping-Transformer
A specialized schema translation transformer that maps heterogeneous source schemas to valid FHIR JSON layouts.
HL7 FHIR Validator CLI
The official validation engine used to confirm that translated JSON files adhere strictly to specified FHIR profiles.
Expected Outcome:
An automated extract-transform-load (ETL) pipeline that reads tabular health records and produces standard-compliant FHIR bundles ready for exchange.
$ openphr deploy legacy-sql-to-fhir-mapping
Cookbook 29: Real-Time Patient Vital Signs Anomaly Detection🔗
Process streaming cardiorespiratory telemetry datasets (heart rate, respiration rate, SpO2) and run unsupervised models locally to detect acute medical deterioration and trigger immediate alerts.
Vitals-Anomaly-LSTM
A recurrent autoencoder model optimized to forecast expected baseline vital values and flag out-of-bounds variations.
Scikit-Learn & FastAPI Stream
Libraries for feature scaling, running multivariate Isolation Forests, and setting up WebSocket servers for telemetry streaming.
Expected Outcome:
A lightweight local server script that processes continuous vital sign streams and returns classification results with latency under 5ms.
$ openphr deploy real-time-vitals-anomaly-detection
Cookbook 30: Multi-Modal Clinical RAG for Integrated EHR Ingestion🔗
Ingest unstructured physician progress notes, lab result PDFs, and imaging metadata into a private multi-modal vector database (ChromaDB + LlamaIndex) for instant contextual clinical retrieval.
BioClinical-Embeddings-v2
Dense clinical embedding model optimized for semantic retrieval across medical literature and complex EHR narratives.
LlamaIndex & ChromaDB
Open-source data frameworks and persistent vector stores for building HIPAA-compliant, locally hosted RAG pipelines.
Expected Outcome:
A fully private, offline-capable clinical search engine that indexes multi-modal patient charts and responds to semantic queries with citation sources.
$ openphr deploy multimodal-clinical-rag
Cookbook 31: Automated Clinical Trial Matching via Genomic & Phenotypic Profiling🔗
Match patient electronic health records (genomic VCF variant files, ICD-10 diagnostic codes, lab ranges) against active ClinicalTrials.gov eligibility criteria using local rule-reasoning models.
ClinicalTrial-Matcher-v1
A specialized clinical logic engine designed to evaluate complex inclusion/exclusion criteria against patient phenotypes.
ClinicalTrials.gov APIv2 & PyVCF
Data access clients for querying active trial registries and parsing patient genomic variant files locally.
Expected Outcome:
An automated patient-trial matching pipeline that produces ranked lists of recruiting clinical trials with structured eligibility scores.
$ openphr deploy clinical-trial-matching
Cookbook 32: Real-Time Sepsis Early Warning System (EWS) via Streaming Telemetry🔗
Ingest continuous lab measurements, vitals, and organ system markers to calculate automated SOFA/SIRS scores and trigger early sepsis risk alerts using local gradient-boosted decision trees.
Sepsis-Risk-XGBoost-v2
Calibrated gradient boosting model trained on MIMIC-IV and eICU bedside telemetry for early detection of septic shock.
PySOFA & Scikit-Learn
Python scoring modules for continuous SOFA/qSOFA calculations and real-time risk classification.
Expected Outcome:
A local real-time clinical alerting pipeline that computes organ dysfunction trajectories and triggers proactive clinician notifications up to 6 hours prior to clinical onset.
$ openphr deploy sepsis-early-warning-system
Cookbook 33: Automated Polypharmacy & Adverse Drug Interaction (DDI) Screening🔗
Evaluate complex patient multi-drug regimens against local RxNorm knowledge graphs and ChEMBL bioactivity databases to detect contraindications and adverse drug-drug interactions.
DrugInteraction-BERT-v1
A specialized transformer model trained on pharmacological interaction graphs for classifying multi-drug kinetic conflicts.
ChEMBL API & RxNorm Graph
Local knowledge graph parsers for resolving drug ingredients, mechanism of action, and metabolic pathways.
Expected Outcome:
An offline clinical safety module that audits patient medication lists and flags severe contraindications with mechanistic explanations.
$ openphr deploy polypharmacy-ddi-screening
Cookbook 34: Automated Rare Disease Phenotype Matching via HPO & OMIM🔗
Map uncurated clinical symptom narratives to Human Phenotype Ontology (HPO) terms and compute semantic similarity scores against OMIM & ORPHANET disease models to identify rare genetic syndromes locally.
PhenoBERT-v2
A clinical NER transformer fine-tuned to extract standardized Human Phenotype Ontology (HPO) codes from unstructured EHR notes.
PyHPO & EMBL-EBI OLS
Ontology lookup clients for computing information-theoretic Resnik & Lin similarity metrics against OMIM disease knowledge graphs.
Expected Outcome:
A local diagnostic decision support pipeline that converts patient clinical notes into standardized HPO profiles and ranks rare genetic disease candidate matches.
$ openphr deploy rare-disease-phenotype-matching
Cookbook 35: Longitudinal Cancer Biomarker & Liquid Biopsy Tracking🔗
Monitor circulating tumor DNA (ctDNA) kinetics, variant allele fractions (VAF), and serum protein markers across treatment regimens using local Bayesian trajectory models to detect early disease recurrence.
ctDNA-Recurrence-Bayes-v1
A Bayesian longitudinal modeling framework calibrated to detect molecular residual disease (MRD) relapse from NGS liquid biopsy panels.
BioPython & PyVCF
Sequence analysis and variant call format parsers for processing serial cfDNA sequencing assays and serial tumor marker feeds.
Expected Outcome:
An offline oncology decision support module that visualizes clonal evolution curves and alerts clinicians to minimal residual disease (MRD) relapse prior to imaging.
$ openphr deploy oncology-liquid-biopsy-tracking
Cookbook 36: Automated Clinical Trial Eligibility Matching via Genomic & EHR Structuring🔗
Automatically match patient EHR records and genomic variant profiles (VCF) against complex inclusion/exclusion criteria from ClinicalTrials.gov to accelerate clinical trial recruitment locally.
TrialMatch-Clinical-v1
A clinical logic encoder fine-tuned to extract eligibility rules from unstructured clinical trial protocols and evaluate patient vector matches.
ClinicalTrials.gov API & PyVCF
API client and variant call format parsers for fetching active trial protocols and cross-referencing somatic/germline genomic variants.
Expected Outcome:
A local recruitment decision support tool that ranks relevant recruiting clinical trials for patients based on molecular biomarkers and disease staging.
$ openphr deploy clinical-trial-eligibility-matcher
Cookbook 37: Real-Time Sepsis Early Warning & Hemodynamic Instability Prediction🔗
Ingest continuous ICU bedside vital telemetry and laboratory feeds to predict organ dysfunction and septic shock onset up to 6 hours before clinical deterioration using local recurrent attention models.
SepsisPredict-Recurrent-v1
A temporal attention network trained on MIMIC-IV time-series telemetry to detect subtle subclinical hemodynamic instability and systemic inflammation.
MIMIC-IV Benchmark & PyHealth
Clinical time-series modeling framework for parsing real-time arterial pressure, heart rate variability, and lactate biomarker trends.
Expected Outcome:
An offline ICU clinical decision support monitor that computes SOFA/qSOFA scores and alerts care teams to early hemodynamic decompensation.
$ openphr deploy sepsis-early-warning
Cookbook 38: Multi-Omic Polygenic Risk Score (PRS) & Disease Susceptibility Pipeline🔗
Calculate individual genome-wide polygenic risk scores across cardiovascular, metabolic, and neurodegenerative phenotypes by scoring patient VCF files against GWAS catalog effect weights offline.
PRS-Score-Clinical-v1
A calibrated polygenic risk scoring engine implementing clumping and thresholding (C+T) and Bayesian LDpred2 algorithms for disease risk estimation.
GWAS Catalog & PLINK2 Engine
Genomic data processing suite for aligning patient genotype dosages against published summary statistics and linkage disequilibrium references.
Expected Outcome:
An offline clinical genomics pipeline that outputs quantile-normalized disease susceptibility percentiles and actionable screening recommendations.
$ openphr deploy polygenic-risk-scoring
Cookbook 39: Automated Psychiatric Progress Note Summarization & DSM-5 Symptom Extraction🔗
Structure unstructured clinical mental health progress notes into standardized DSM-5 diagnostic criteria and compute longitudinal patient sentiment and symptom trajectory trends offline.
PsychNote-NLP-v1
A domain-adapted transformer model for identifying affective status, suicidal ideation indicators, and psychotic features from psychiatric narratives.
DSM-5 Taxonomy & MedSAM-Text
Clinical NLP pipeline for mapping clinical terminology to DSM-5 codes and generating FHIR-compliant DiagnosticReport bundles.
Expected Outcome:
An offline mental health clinical decision support engine that structures longitudinal psychiatric progress notes into actionable symptom timelines.
$ openphr deploy psychiatric-note-summarizer
Cookbook 40: Pediatric Asthma & Allergy Exacerbation Forecasting from Environmental & EHR Data🔗
Predict 7-day asthma and allergic rhinitis exacerbation risk for pediatric patients by integrating personal EHR allergy profiles with real-time hyper-local air quality (AQI) and pollen sensor streams offline.
PediatricAsthma-Forecaster-v1
A temporal gradient boosted tree ensemble trained on pediatric respiratory telemetry and environmental exposure correlations.
OpenAQ API & PollenStream-Local
Environmental data ingestion engine for fetching local PM2.5, ozone, and aeroallergen counts without disclosing patient location coordinates.
Expected Outcome:
An offline pediatric environmental risk engine that outputs localized allergen threshold alerts and preventive medication reminders.
$ openphr deploy pediatric-asthma-forecaster
Cookbook 41: Multi-Organ Systems Biology Network Analysis & Drug Repurposing Candidates🔗
Map complex multi-organ interactomes using single-cell transcriptomic profiles and graph neural networks to identify patient-specific drug repurposing candidates and off-target side effect risks offline.
OrganInteractome-GNN-v1
A heterogeneous graph neural network trained on tissue-specific protein-protein interaction networks and drug target interactomes.
STRING-DB & BioNetEngine-Local
Local graph analysis library for propagating transcriptomic signals across multi-organ pathways to uncover repurposed therapy targets.
Expected Outcome:
An offline network medicine platform that ranks high-confidence repurposed compounds and pathway perturbation targets for refractory chronic conditions.
$ openphr deploy organ-interactome-analyzer
Cookbook 42: Automated Rare Disease Variant Prioritization & Phenotype-Genotype Matching🔗
Rapidly diagnose undiagnosed rare Mendelian conditions by matching patient HPO (Human Phenotype Ontology) clinical features against whole-genome VCF variants using local LLM-assisted variant scoring models offline.
RareVariant-Ranker-v1
A pheno-genomic variant prioritizing neural network trained on ClinVar annotations and Human Phenotype Ontology semantic embeddings.
ClinVar & HPO-Sim-Local
Offline genomic annotation engine for calculating semantic phenotypic similarity and extracting ACMG pathogenicity criteria.
Expected Outcome:
An offline clinical genetics diagnostic pipeline that ranks candidate pathogenic variants alongside supporting ACMG evidence and gene-phenotype similarity scores.
$ openphr deploy rare-variant-prioritizer
Cookbook 43: Offline Pharmacogenomic Risk Stratification & Drug-Gene Interaction Checker🔗
Cross-reference patient star-allele genotypes (CYP2D6, CYP2C19, TPMT, DPYD) against CPIC clinical guidelines to predict adverse drug reaction risks and optimize medication dosing offline.
PGx-Predictor-v1
A clinical metabolizer phenotype classifier that maps diplotypes to actionable CPIC dosage recommendations.
CPIC-DB & PharmGKB-Local
Local pharmacogenomic rules engine for evaluating drug-gene interactions and high-risk prescription warnings offline.
Expected Outcome:
An offline pharmacogenomic clinical decision support engine that flags high-risk drug-gene interactions and generates personalized alternative medication recommendations.
$ openphr deploy pgx-risk-stratifier
Cookbook 44: Local Multimodal Dermatology AI for Lesion Risk Assessment🔗
Analyze dermoscopic images alongside patient risk factors (skin type, family history, UV exposure) using on-device vision-language models to triage skin lesions offline.
DermVision-VLM-v1
A compact multimodal vision-language transformer fine-tuned on dermoscopic image benchmarks for malignancy risk stratification.
ISIC-Archive & HAM10000-Local
Offline dermoscopic reference database for calculating visual similarity metrics across pigmented skin lesions.
Expected Outcome:
An offline clinical dermatology triaging pipeline that provides differential diagnoses for skin lesions with confidence scores and privacy-preserving risk reports.
$ openphr deploy derm-lesion-triage
Offline Pediatric Growth & Developmental Milestone Tracker
Track pediatric growth percentiles (height, weight, head circumference) and developmental milestones using offline CDC/WHO growth standards and local anomaly detection models to flag developmental delays.
Required Components:
GrowthCurve-ML-v1
Non-parametric quantile regression model trained on pediatric anthropometric data to identify growth velocity anomalies.
CDC-WHO-Percentiles & Denver-II-Local
Local lookup database containing CDC & WHO Z-scores alongside Denver II developmental milestone thresholds.
Expected Outcome:
A local pediatric decision support tool that calculates exact Z-scores, plots growth trajectories, and alerts clinicians to early developmental milestone deviations.
$ openphr deploy pediatric-growth-tracker
Offline Ophthalmology Retinal OCT Layer Segmentation & Biomarker Quantifier
Segment macular OCT B-scans offline to quantify central subfield thickness, intraretinal fluid (IRF), and subretinal fluid (SRF) for longitudinal monitoring of diabetic macular edema (DME) and wet AMD.
Required Components:
RetinaOCT-Segment-v1
Deep convolutional network trained on high-resolution OCT volumes for 9-layer retinal segmentation and fluid volumetric analysis.
DUKE-OCT-Archive & ARVO-Biomarker-DB-Local
Offline retinal imaging reference database and clinical biomarker threshold definitions.
Expected Outcome:
An offline ophthalmology decision support pipeline generating automated retinal layer thickness maps and fluid volume metrics for anti-VEGF response tracking.
$ openphr deploy oct-retina-quantifier
Offline Renal Function & Acute Kidney Injury (AKI) Predictive Trajectory Model
Track offline serum creatinine trajectories, urine output, and nephrotoxic drug exposures to forecast KDIGO stage 1-3 Acute Kidney Injury 48 hours prior to clinical onset.
Required Components:
RenalRisk-RNN-v1
Recurrent neural network trained on longitudinal EHR lab trends and physiological telemetry for early AKI trajectory forecasting.
KDIGO-Guidelines-Local & MIMIC-Renal-DB
Offline KDIGO diagnostic criteria database and nephrotoxicity risk scoring matrix.
Expected Outcome:
An offline nephrology decision support tool alerting clinicians to impending AKI risk and recommending nephrotoxic medication dose adjustments before renal impairment progresses.
$ openphr deploy renal-aki-forecaster
Offline Sepsis Early Warning & Systemic Inflammatory Response Syndrome (SIRS) Monitor
Continuously compute offline SOFA scores and monitor vitals telemetry (heart rate, temperature, lactate, MAP) to detect septic shock onset up to 6 hours prior to clinical recognition.
Required Components:
SepsisAlert-LSTM-v1
High-frequency time-series LSTM trained on ICU physiological streams for zero-latency sepsis risk stratification.
SOFA-Score-Engine & MIMIC-Sepsis-DB-Local
Offline Sequential Organ Failure Assessment calculator and clinical bundle adherence checklist.
Expected Outcome:
An offline critical care alert engine that triggers early sepsis warnings, calculates continuous organ failure scores, and recommends rapid antibiotic bundle execution without cloud dependencies.
$ openphr deploy sepsis-early-warning
Offline Pulmonology Asthma & COPD Exacerbation Early Warning System
Process continuous respiratory acoustic telemetry (wheezing, crackles) and digital spirometry flows offline to predict acute COPD and asthma exacerbation risk up to 72 hours in advance.
Required Components:
RespTrack-WavLM-v1
Self-supervised audio transformer fine-tuned for high-accuracy adventitious lung sound classification and wheeze detection.
Acoustic-Stethoscope-Analyzer & Spirometry-DB-Local
Offline peak flow analyzer, GOLD COPD staging calculator, and rescue inhaler usage tracker.
Expected Outcome:
An edge-ready respiratory monitoring agent that alerts patients and care teams to impending pulmonary exacerbations, recommending timely steroid or bronchodilator adjustments before emergency hospitalization occurs.
$ openphr deploy respiratory-exacerbation-monitor
Offline Gastroenterology IBD & Crohn's Disease Flare Predictor
Combine longitudinal fecal calprotectin assays, abdominal ultrasound telemetry, and patient-reported outcome logs offline to forecast Inflammatory Bowel Disease (IBD) mucosal flare-ups 14 days in advance.
Required Components:
GutFlare-MultiModal-v1
Multimodal transformer predicting subclinical mucosal inflammation using biomarker trends and gastrointestinal motility features.
Calprotectin-Tracker & Endoscopy-DB-Local
Offline fecal calprotectin quantifier, Mayo Endoscopic Score calculator, and biologic therapeutic level logger.
Expected Outcome:
An offline gastroenterology decision-support model that detects active mucosal inflammation before clinical relapse, enabling proactive biologic dose optimization and preventing acute disease flares.
$ openphr deploy ibd-flare-predictor
Offline Hematology Anemia & Hemoglobin Trajectory Predictor
Track complete blood count (CBC) telemetry, reticulocyte indices, and iron panel dynamics offline to forecast chronic disease anemia trajectories and erythropoietin dosing requirements.
Required Components:
HemoTrack-RNN-v1
Recurrent deep sequence model forecasting longitudinal hemoglobin decay rates and red blood cell survival dynamics.
CBC-Differential-Tracker & Iron-Panel-DB-Local
Offline ferritin/transferrin calculator, transferrin saturation (TSAT) index analyzer, and erythropoiesis-stimulating agent (ESA) logger.
Expected Outcome:
An edge hematology tracking system that forecasts severe anemia events weeks before clinical decompensation, enabling precision iron supplementation and ESA titration without cloud dependency.
$ openphr deploy anemia-trajectory-predictor
Offline Nephrology CKD Stage Progression & eGFR Trajectory Predictor
Track serial serum creatinine, blood urea nitrogen (BUN), urine albumin-to-creatinine ratio (uACR), and systemic blood pressure logs offline to forecast 90-day eGFR trajectories and chronic kidney disease stage transitions.
Required Components:
RenalTrack-LSTM-v1
Longitudinal temporal neural network estimating renal function loss slopes and acute-on-chronic renal failure risks.
uACR-Calculator & CKD-EPI-DB-Local
Offline CKD-EPI 2021 equation engine, proteinuria quantifier, and nephrotoxic drug exposure monitor.
Expected Outcome:
An edge nephrology predictive suite that detects accelerated kidney function loss and alerts clinical teams to stage progression risks without sending sensitive lab values to cloud servers.
$ openphr deploy ckd-trajectory-predictor
Offline Ophthalmology Glaucoma & Retinal OCT Analyzer
Process optical coherence tomography (OCT) retinal scans locally to quantify retinal nerve fiber layer (RNFL) thickness, detect diabetic macular edema (DME), and calculate cup-to-disc ratio (CDR) progression.
Required Components:
OphthaVision-ViT-v1
Vision Transformer fine-tuned on 100k+ anonymized retinal OCT B-scans for volumetric macular & optic disc segmentation.
RNFL-Quantifier & DME-Severity-Scale
Offline retinal layer boundary detector, optic cup/disc ratio estimator, and diabetic retinopathy grading engine.
Expected Outcome:
An edge ophthalmology diagnostic workstation that flags early glaucomatous structural loss and macular thickness changes directly in clinic without cloud image transfer.
$ openphr deploy ophthalmology-oct-analyzer
Offline Rheumatology Lupus & Autoimmune Antibody Flare Predictor
Evaluate serial anti-dsDNA antibody titers, complement C3/C4 consumption, and urine protein trajectories to predict systemic lupus erythematosus (SLE) organ flares and calculate SLEDAI-2K disease activity on-device.
Required Components:
LupusTrack-GNN-v1
Graph Neural Network trained on multi-epitope autoimmune panels, anti-dsDNA kinetics, and complement consumption patterns.
SLEDAI-2K-Calculator & Anti-dsDNA-Quantifier
Offline clinical score calculator, complement C3/C4 monitor, and lupus nephritis renal risk stratifier.
Expected Outcome:
An offline rheumatology decision support system that alerts clinicians to impending lupus flares 30–60 days in advance without sending laboratory records across external networks.
$ openphr deploy lupus-flare-predictor
Offline Rheumatology Autoimmune Arthritis & Biologic Response Predictor
Model joint synovitis telemetry, power Doppler ultrasound scores, anti-CCP titers, and inflammatory markers offline with RheumaTrack-Graph-v1 to predict 12-week DAS28 disease activity trajectories and TNF-alpha biologic treatment response.
Required Components:
RheumaTrack-Graph-v1
Graph neural network for longitudinal autoimmune arthritis trajectory modeling.
Autoimmune-Joint-DB-Local
Local database of de-identified rheumatoid joint exams, DAS28 scores, and bDMARD outcomes.
DAS28-Calculator-CLI
Local CLI utility for computing DAS28-CRP and DAS28-ESR scores with disease activity stratification.
Execution Example:
openphr-cli run rheumatrack-graph-v1 --input-joint-map ./joint_exam.json --serology ./anti_ccp.json --output-das28-prediction ./das28_trajectory.jsonOffline Neurology Epilepsy & Continuous EEG Seizure Prediction Engine
Process multi-channel continuous scalp EEG telemetry locally to detect epileptiform spike-wave discharges, predict onset of clinical seizures 30–60 minutes in advance, and quantify Time-in-Euthymia.
Required Components:
NeuroGuard-EEG-Transformer-v1
Spatial-temporal vision/signal transformer fine-tuned on 50,000+ hours of continuous clinical scalp EEG telemetry for automated seizure onset forecasting.
Epileptiform-Spike-Detector & Seizure-Risk-Index
Offline real-time spectral analyzer, muscle/blink artifact remover, and automated seizure risk index generator.
Expected Outcome:
An offline neurology monitoring system that alerts clinical care teams and patients to impending epileptic seizures 30–60 minutes in advance without transmitting high-bandwidth raw neural signals to cloud services.
$ openphr deploy eeg-seizure-predictor
Offline Critical Care Mechanical Ventilation & ARDS Trajectory Predictor
Analyze continuous mechanical ventilator waveform telemetry (P-V loops, airway pressure, PaO2/FiO2 ratio, driving pressure) to forecast ARDS severity trajectories, flag barotrauma risk, and optimize extubation readiness on-device.
Required Components:
VentGuard-Transformer-v1
High-frequency waveform transformer model fine-tuned on ICU mechanical ventilation telemetry and arterial blood gas (ABG) panels.
ARDS-Severity-Index & Extubation-Readiness-Score
Offline PaO2/FiO2 ratio trend calculator, driving pressure monitoring suite, and automated weaning trial predictor.
Expected Outcome:
An edge ICU ventilator decision support system that alerts intensivist teams to weaning opportunities and impending lung injury without cloud dependencies.
$ openphr deploy ventilator-ards-predictor
Offline Hepatology Cirrhosis & Portal Hypertension Risk Predictor
Analyze longitudinal liver function laboratory panels (bilirubin, INR, creatinine, sodium), transient elastography (FibroScan) readings, and abdominal ultrasound findings to forecast 90-day decompensation risk and MELD-Na trajectory on-device.
Required Components:
HepatoRisk-GNN-v1
Graph neural network fine-tuned on multi-center liver transplant registry datasets for predicting hepatic decompensation and variceal bleeding risk.
MELD-Na-Calculator & Fibrosis-Stage-Analyzer
Offline clinical score engine, transient elastography parser, and hepatic encephalopathy grading assistant.
Expected Outcome:
An edge hepatology clinical decision support system that alerts gastroenterologists to impending decompensation and esophageal variceal risk without sending patient laboratory data to external clouds.
$ openphr deploy hepatology-cirrhosis-predictor
Analyze peripheral blood smear microscopic images, automated CBC differential parameters (MCV, RDW, reticulocyte count), and bone marrow aspirate cytomorphology to classify refractory anemias and MDS (myelodysplastic syndrome) risk on-device.
Required Components:
HematoVision-CNN-v1
Vision Transformer and CNN ensemble fine-tuned on peripheral blood smear microphotographs and bone marrow aspirate cell morphometry.
Hematology-Blood-DB-Local & MDS-Risk-CLI
Offline cell differential index engine, iron deficiency vs. thalassemia differentiator, and IPSS-R risk scoring tool.
Expected Outcome:
An edge hematology decision support pipeline that assists clinical hematologists in identifying dysplastic cell lines and refractory anemias directly at the microscope.
$ openphr deploy hematology-anemia-classifier
Analyze optical coherence tomography (OCT) cross-sectional B-scans and fundus photographs to automatically segment drusen volume, fluid accumulation (SRF/IRF), and age-related macular degeneration (AMD) staging on-device.
Required Components:
RetinaNet-OCT-v1
3D Convolutional and Transformer segmentation model fine-tuned on retinal OCT volume scans and color fundus photographs.
Ophthalmology-Retina-DB-Local & AMD-Staging-CLI
Offline retinal layer boundary parser, subretinal fluid quantifier, and AREDS2 clinical score classifier.
Expected Outcome:
An edge ophthalmic AI diagnostic system that assists retina specialists in staging macular degeneration and tracking anti-VEGF treatment response directly on clinic imaging workstations.
$ openphr deploy ophthalmic-retina-engine