AI Diagnostic Assistant
HealthcareComputer VisionRegulatedMobile

AI Diagnostic Assistant

Assisting radiologists in detecting early signs of anomalies in X-rays with Computer Vision.

Role

Product Lead

Timeline

24 Months

Team

4 Research Scientists, 3 Engineers, Clinical Partners

Context & Problem

The Context

Radiologists are overworked, leading to burnout and potential diagnostic errors.

The Problem

High volume of scans means less time per patient. Subtle anomalies are easily missed in early stages.

Opportunity & Vision

"An AI 'second pair of eyes' that highlights potential regions of interest, prioritizing the worklist for radiologists."

Data & AI Opportunities

Convolutional Neural Networks (CNNs) trained on large datasets of annotated X-rays to detect specific pathologies.

PyTorchDICOMEdge AIiPadOS

System Architecture

Edge deployment for privacy and speed. DICOM integration. FDA clearance workflow built into the development process.

Data Ingestion

Kafka / IoT Sensors

AI Processing

TensorFlow / Spark

API Layer

FastAPI / GraphQL

Client App

React / Mobile

End-to-End Encrypted & HIPAA Compliant

Tradeoffs & Prioritization

We focused on high sensitivity (recall) to ensure no potential issues were missed, accepting a higher false positive rate which the radiologist could easily dismiss.

Execution & Collaboration

Partnered with 3 major hospitals for data and validation. Navigated complex regulatory landscape (HIPAA, FDA).

Metrics & Impact

Improved detection rate of early-stage nodules by 20%. Reduced average review time per scan by 30%.

40%

Reduction

In unplanned downtime

$5M

ARR

Generated in 18 months

95%

Accuracy

On critical failure modes

Lessons Learned

"User trust is paramount in healthcare. Explainability (Grad-CAM heatmaps) was just as important as model accuracy."

Abdoulaye Bah | AI Product Manager Portfolio