Book a demo
Book a demo

AI-Powered Imaging for Cancer Detection

AI that helps clinicians see what deserves a second look. OncoDx analyzes medical images to highlight suspicious findings and support informed clinical decisions.

Book a demo Contact us
The Science

How AI Interprets Medical Images

Medical images contain patterns in tissue structure, shape and appearance. Some findings can be subtle or difficult to distinguish from surrounding anatomy.

OncoDx uses deep learning models trained for specific imaging tasks to identify suspicious regions. Depending on the application, it analyzes CT volumes across multiple slices or processes endoscopy images and video to detect potential lesions.

The findings are presented as visual markers or segmentation overlays, giving clinicians additional information to review alongside the original images and the patient’s clinical context.

How It Works

From Medical Images to Clinical Review

Hundreds of CT slices. Continuous endoscopy footage. Subtle findings within complex anatomy. OncoDx adds a focused layer of AI analysis, helping clinicians identify areas that warrant closer review.

1. Image input
1. Image input

Upload a supported CT study, endoscopy image or video, or receive input through a configured video connection.

2. AI analysis
2. AI analysis

A specialized AI model analyzes the images and marks suspected regions for closer inspection.

3. Clinical review
3. Clinical review

The clinician examines AI findings alongside the original images and clinical information to determine the next step.

Measured, not generalized

Evidence behind
each application.

Results reported for specific models and evaluation settings.

92.38%Nodule-level sensitivity
98.31%Case-level sensitivity
0.432False positives per CT scan
2.3 minAverage inference time per case

LUNA16 held-out test set: 88 CT scans, 105 reference nodules. Nodule-level sensitivity 95% CI: 87.18–97.08%. Case-level sensitivity means at least one reference nodule was correctly detected; it does not imply all nodules were found.

Research results are not a platform-wide accuracy claim or a guarantee of clinical performance. Results depend on input data, imaging protocols, hardware and deployment conditions. Site-specific evaluation is required.
Specialized by design

Four cancer applications.
A growing field of vision.

Different organs. Different imaging challenges. Dedicated models developed for the anatomy and task at hand.

CHEST CT / RESEARCH EXAMPLE
3D CHEST CT

A second look at pulmonary nodules.

Analyze the full CT volume to flag suspected pulmonary nodules for radiologist review.

  • DICOM series input and volumetric 3D analysis
  • Localized regions with model confidence scores
  • Lung-field filtering to reduce out-of-lung predictions

A detected nodule is not a diagnosis of lung cancer.

Dr. Nguyen Duc Phuong

“Our goal with OncoDx is to help clinicians detect even small nodules and subtle abnormalities. We combine AI research with specialized medical image analysis to make suspicious findings easier to locate and review, supporting informed clinical judgment and guiding further assessment.”

Dr. Nguyen Duc Phuong CEO & Head of AI R&D Lab
Explore OncoDx in Action

Discover how AI highlights suspicious findings in medical images. Book a demo

FAQ

Frequently asked questions about our product.

OncoDx is an AI medical imaging solution developed by DXP Software. Part of AI Imaging within the ClinicOS ecosystem, it analyzes medical images to highlight suspicious findings and support clinical review.

OncoDx focuses on four application areas: lung, pancreatic, colorectal and gastric cancer. Capabilities and availability vary by application and model version, with additional applications under development.

Documented applications include chest CT for pulmonary nodule detection, contrast-enhanced abdominal CT for pancreatic analysis, and colorectal endoscopy images and video for polyp detection. Specific input requirements are defined for each model.

OncoDx highlights suspected regions so clinicians can examine them more closely. Depending on the application, outputs may include bounding boxes, segmentation overlays or model confidence scores, reviewed alongside the original images.

No. A highlighted region, detected nodule or polyp does not confirm cancer. Clinicians interpret the findings alongside medical history, other examinations and, where appropriate, pathology results.

Yes. The lung CT model analyzes volumetric data across multiple slices. The colorectal application processes images and video to support polyp detection. Processing methods and performance depend on the selected model and system configuration.

Each model is evaluated for its specific task using relevant measures, such as sensitivity, specificity, false positives, segmentation quality and processing time. Published results apply to the stated model and test conditions; performance should also be assessed in the intended deployment setting.

A confidence score indicates how strongly the model supports a particular prediction. It is not the patient’s probability of having cancer and should be interpreted together with the image findings and clinical context.

DXP assesses your imaging equipment, data formats, software and infrastructure to determine a suitable integration approach. Connectivity with ClinicOS or other clinical systems is defined according to the project’s requirements and supported interfaces.

Contact DXP to arrange a demonstration or discuss a structured evaluation. Together, we can define the application, permitted data use, technical requirements and evaluation criteria appropriate for your institution.