AI-based analysis of radiology images
A deep-learning model scans the image tile by tile, highlights a region that may need attention with a heat map and a contour, and moves the study up the worklist. The system is decision support; assessment and diagnosis remain with the specialist.
Region score, size and position are written to the study record; a study above the threshold is moved up the worklist and presented to the specialist with the heat-mapped image. The final assessment and diagnosis belong to the specialist.
- 01 Image
- 02 AI analysis
- 03 Heat map
- 04 Routing
The study is fetched from the archive and loaded on the diagnostic monitor.
- Region score
- —
- Region size
- —
- List position
- —
Scenes are illustrative simulations; they are not customer references or performance guarantees.
Need
In the grey-scale image the region of interest differs from its surroundings by only a few tones and has an irregular shape; noise, imaging parameters and device differences make the same finding look different. A false alarm costs the specialist time, a missed finding affects the patient.
Solution approach
The study is loaded from the archive and the model scans the image tile by tile, producing a score per tile; the row scores are plotted against the threshold line. A region above the threshold is shown on the image with a heat map and a contour, with its score and size. The result is written to the worklist; the model only prioritises, the decision belongs to the specialist.
Process steps
- 01
Image
The study is fetched from the archive and loaded on the diagnostic monitor.
- 02
AI analysis
The model scans the image tile by tile; each row score is plotted against the threshold line.
- 03
Heat map
A region above the threshold is shown on the image as a heat map with a contour.
- 04
Routing
The result is written to the worklist; the final assessment is the specialist’s.
Related applications
Let us assess your application together
We test your sample in our laboratory and report on resolution, optics and lighting selection.