Medical Image Analysis

AI analyzes medical images to detect, classify, segment, and quantify findings for clinical use.

Business impact

• Accuracy — One skin-lesion system reported 86% overall accuracy.
• Sensitivity — One skin-lesion system reported 88% sensitivity for melanoma and 92% for basal cell carcinoma.
• Correlation with pathologist scoring — Ki67 image-analysis scores correlated with pathologists at r=0.95, and r=0.98 for lymph node scoring.
• Response time — One dermatology deployment processed studies in under one second.

Data requirements

• Medical images (image) — X-rays, microscopy images, histopathology slides, dermatology images, MRI, CT, ultrasound, and mammography are all represented in the evidence.
• Clinical data (tabular) — One skin-lesion system combined images with clinical data.
• Pathologist scores (tabular) — Used as comparison data for automated Ki67 assessment.
• Annotated datasets (image) — Training and evaluation used labeled medical image datasets and specific histopathology datasets.

AI methods and techniques

• Image classification — Assigns images or regions to disease or tissue categories.
• Segmentation — Separates cells, tissues, or structures for measurement.
• Detection/localization — Finds abnormalities or suspicious regions in scans.
• Quantification — Produces measurements and scores from image data.
• Deep learning — Used across the image-analysis systems described for medical imaging tasks.

AI models and model families

CNN, Graph Transformers, CLAM, neural network

Industries

Real-world evidence

8 documented case studies on record.

Companies using this: SimonMed Imaging, Center for Label-free Imaging and Multiscale Biophotonics, Sant Joan d'Alacant University Hospital, Sartorius, MobiDev, University of São Paulo (USP), Ribeirão Preto campus, University of Cologne, Cerba Research.