ML Algorithms to Identify Dental Fraud with 95% Accuracy
About Our Client
The Client is an innovative dental care insurance management startup.
Preventing Dental Fraud with Machine Learning
The Client is developing a cutting-edge software product to help identify insurance errors and dental fraud. Currently, dental care providers sometimes make mistakes when filling out insurance claims (e.g., listing a wrong tooth code). There are also cases of intentionally changing the diagnosis and submitting claims with the same or manually altered X-rays multiple times to receive higher reimbursement.
The Client’s software product is meant to use computer vision to analyze reimbursement claims and attached X-rays to identify the dental condition and compare it with the diagnosis and procedures listed in the claim. In addition, the software is to compare the X-ray image with images uploaded to the software database, identify any duplicates or manual image modifications, and report fraud.
The Client needed to augment its medical software development team with a data scientist experienced in computer vision to help build reliable ML algorithms for oral health condition identification and medical image comparison. The Client was looking for a trustworthy vendor that could promptly provide a fitting specialist and contribute to a faster software product launch.
Rapid Team Augmentation with a Computer Vision Expert
Being impressed with ScienceSoft’s experience in medical image analysis, the Client turned to us to augment its team with a skilled machine learning engineer. To meet the Client’s needs, ScienceSoft allocated a full-time Senior Data Scientist/Senior Python Developer with team lead experience and deep knowledge of healthcare specifics. It was decided to use T&M (time and material) pricing model to provide for service scalability.
Within a week, ScienceSoft’s developer was ready to join the project. After a short knowledge transfer period and project documentation analysis, he started working on computer vision ML algorithms along with the Client’s data scientists.
The joint team was to build algorithms for:
- Tooth type (incisor, canine, premolar, molar) and location identification (coded using a standard tooth numbering system).
- Automatic detection of dental problem location within a tooth.
- Identification of the oral health condition type (cavity, decay beneath the existing filling, abscess, root rot, cyst, etc.).
- Comparison of the medical images with images from the database to validate the authenticity of an X-ray scan.
- Identification of duplicated medical images with minor changes (e.g., image contrast tuning, adding new image elements using editing programs, image alignment changes) to prevent fraudulent practices used by dental care providers.
ScienceSoft’s expert successfully participated in building deep learning models and leveraged his 10+ years of experience to overcome technical challenges along the way. For example, he contributed to harmonizing different radiology information systems’ (RISs) formats of X-ray files. The project team also tackled the issues with X-rays quality, sizing, alignment, irrelevant information mentioned on the scan, and more.
ScienceSoft, as an ISO 13485-certified IT services provider, followed the quality management approach detailed in the standard, as the product was considered SaMD (software as a medical device).
Achieving 95% Algorithm Accuracy
After the 6-month-long engagement of ScienceSoft’s Senior Data Scientist, the Client got the market-ready software MVP. On average, the medical image recognition algorithms demonstrated 95% accuracy. After the development is finished, the Client intends to submit the product to the FDA for clearance.
The Client’s in-house development team highly appreciated the ML expertise of ScienceSoft’s engineer, his proactive approach, and impeccable communication skills.
Technologies
Python, Pytorch.
