DICOM Image Generation Module to Visualize AI Findings

About the Customer

The Customer is a global provider of AI-powered medical imaging solutions for various medical specialties.

The Need to Visualize AI Diagnostic Findings

One of the Customer’s popular products is AI software for X-ray and MRI image analysis and smart disease diagnostics. By uploading X-ray and MRI scans to the software, medical professionals can receive a fast diagnostic analysis of the patient’s condition. However, in its initial version, the software only displayed the AI findings as plain text. To make the product more convenient for medical staff, the Customer wanted its software to display the AI output directly on the uploaded image and contour the abnormal area.

To achieve this, the Customer needed to build a new software module that would generate DICOM images with overlayed AI output based on the uploaded medical images. The company sought an ISO 13485 and IEC 62304-compliant healthcare software development vendor to scale up the in-house team with senior Python developers and QA specialists.

DICOM Image Generation Module

Having 18 years of experience in healthcare IT, 10 years in image analysis, and being ISO 13485-certified, ScienceSoft fully met the Customer’s requirements. We provided two senior Python developers and a senior QA engineer to join the Customer’s team.

The augmented team delivered the new software module in 3 months. The end product works as follows:

• A physician uploads an X-ray or MRI image to the software.
• The AI engine analyzes the medical image and produces the result (in text format).
• The new AI image generation module converts the image from DICOM standard, adds the label with image analysis findings to the medical image and highlights the problematic areas with a colored contour.
• Using DICOM Secondary Capture technology, the AI-generated image is transformed back into DICOM format without losing the initial image quality.

The project team integrated the software module with the Customer’s PACS (picture archiving and communication system) C-STORE client via custom APIs. Now, physicians can send the generated DICOM images straight to their hospitals’ PACS systems without uploading them manually.

Regulatory Compliance and Safe Software Use

As the original software product was popular in the EU, ScienceSoft ensured the new module would fully comply with GDPR requirements.

The Customer’s software product is considered SaMD (software as a medical device), so the new functional module was developed and tested according to ISO 13485 and IEC 62304 standards. ScienceSoft’s team members created the software documentation for the new module in line with the above standards to ensure smooth software submission to a Notified Body designated under MDR.

One of the key project concerns was the potential risk of improper software use and following diagnostic negligence. To address it, ScienceSoft’s team ensured that each generated image is marked with a disclaimer stating that it is not a definitive medical diagnosis but an AI-produced assumption.

Improved User Satisfaction and a New AI Module in 3 Months

Thanks to fast team augmentation with ScienceSoft’s senior developers and a QA engineer, the Customer built a new software module for DICOM image generation in only 3 months. With the improved AI software for medical image analysis, the end users got more detailed interpretations of images for effective patient diagnostics and treatment. The new software module complies with GDPR, ISO 13485, and IEC 62304 requirements. After rolling out the software product update, the Customer reported higher medical staff satisfaction rates.

Technologies and Standards

Python, Pytorch.

DICOM, ISO 13485, IEC 62304.