Discovery and Requirements Gathering: Curate partnered with the healthcare provider’s radiology, IT, and compliance teams to:

• Identify key image modalities (e.g., chest X-rays, MRIs) for automation

• Define diagnostic categories (e.g., tumor detection, pneumonia signs)

• Establish performance benchmarks (precision, recall, latency)

• Ensure compliance with HIPAA and FDA regulatory frameworks

Model Development and System Integration:

• Data Preparation: Curated a dataset of over 1 million de-identified imaging scans, annotated by expert radiologists.

• Model Training: Built convolutional neural networks (CNNs) using transfer learning and custom architectures for modality-specific tasks.

• Use Cases: Included early detection of lung abnormalities, bone fractures, and brain lesions.

• Performance Optimization: Models achieved over 92% precision and recall across key diagnostic targets.

• Workflow Integration: Deployed via REST APIs into the provider’s PACS and EHR systems to surface alerts and classification labels directly to clinicians.

Automation and Quality Assurance:

• Auto-Triage: Flagged high-risk images (e.g., suspected tumors, fractures) for prioritized human review.

• Heatmap Visuals: Used Grad-CAM overlays to provide visual explanations of anomaly zones, aiding radiologist interpretation.

• Continuous Learning: Enabled incremental retraining with newly reviewed images to improve model robustness.

• Redundancy Checks: Maintained manual review protocols for flagged scans to ensure quality assurance.

Stakeholder Engagement & Change Management:

• Radiologist Collaboration: Involved radiologists in validation, feedback, and refinement of model predictions.

• Compliance Oversight: Ensured all image handling and storage adhered to HIPAA and FDA guidelines.

• Training & Adoption: Delivered workshops on AI-assisted workflows and interpretation of AI-generated outputs.

• Phased Rollout: Initiated with select facilities and scaled following positive results and clinician acceptance.