Urine microscopy, reimagined

Fast. Reliable. Compact.

Calm. Precise. Ready for everyday workload.

Thesis

MicroView AI

Leveraging Large Vision-Language Models in an Augmentative Raspberry Pi–Based System for Urine Microscopy Analysis

Supported by

Pamantasan ng Lungsod ng Maynila

College of Engineering

Vision-Language Insight

Explainable AI surfaces critical findings with clear context for medical technologists.

Lab-Ready Experience

Minimal setup through a Raspberry Pi bridge designed for consistent lab workflows.

Confident Reporting

Structured summaries, quality captures, and trend comparisons prepared for review.

Two-Stage AI Analysis Pipeline

From image acquisition to clinical reporting, our workflow uses YOLO for coarse-grain detection followed by Gemini for fine-grain analysis, ensuring both speed and accuracy.

Step 1

Image Acquisition

Capture high-quality microscopy images from the microscope feed for analysis.

Step 2

Coarse-Grain Analysis

YOLO v11 performs initial object detection to identify and locate sediment types with bounding boxes.

Step 3

Fine-Grain Analysis

Gemini 2.5 Pro performs detailed analysis, verifies YOLO detections, and provides clinical context.

Step 4

Report Generation

Generate comprehensive reports with structured findings, annotated images, and clinical recommendations.

Powered by Hybrid AI Technology

Combining the speed of YOLO v11 object detection with the clinical expertise of Google Gemini 2.5 Pro for accurate, explainable urinalysis analysis.

YOLO v11 - Coarse Detection

First-stage analysis: Fast object detection model performs initial coarse-grain identification of sediment types

✓Rapid bounding box generation for detected objects
✓Initial classification of sediment types
✓Confidence scores for each detection
✓Multi-class detection (RBC, WBC, crystals, casts, etc.)

Gemini 2.5 Pro - Fine Analysis

Second-stage analysis: Advanced vision-language model performs detailed fine-grain verification and refinement

✓Verifies and refines YOLO coarse detections
✓Performs comprehensive systematic image scan
✓Provides detailed morphology descriptions
✓Generates clinical context and recommendations

Hybrid Pipeline

Two-stage approach: Coarse detection followed by fine analysis ensures both speed and accuracy

✓YOLO provides fast coarse-grain initial detection
✓Gemini performs fine-grain detailed analysis
✓Error correction and missed detection recovery
✓Explainable AI with comprehensive reasoning

How It Works Together

Image Acquisition

→

YOLO v11 (Coarse)

→

Gemini 2.5 Pro (Fine)

→

Clinical Report

First, YOLO v11 performs coarse-grain analysis to quickly identify and locate sediment types. Then, Gemini 2.5 Pro performs fine-grain analysis to verify detections, refine classifications, and provide detailed clinical context. This two-stage approach ensures both speed and accuracy.

Crafted to elevate daily diagnostics.

MicroView AI streamlines sample validation, keeps care teams in sync, and preserves documentation that meets clinical scrutiny.