Research platform
AI-assisted signal analysis: scope and validation rules
OncoFirm is researching software that interprets fluorescence signals from lateral flow assays. The scope is deliberately narrow, and the validation rules are set before any model is built.
What the software is meant to do
The first task is signal processing: locating test and control lines, subtracting background, flagging invalid runs and converting intensity to concentration with lot-specific calibration. Much of that can be done with fixed, explainable algorithms.
The research question is whether learned models add anything on top, for example in recognizing abnormal flow patterns or in combining several biomarker values into a single risk estimate. That has to be shown on independent data. It is not assumed.
No OncoFirm model currently provides a validated diagnosis or clinical recommendation. Clinical interpretation remains with qualified healthcare professionals.
Validation principles we intend to apply
These principles follow published reporting guidance and regulators’ principles for machine learning in medical devices. Each study protocol will set the binding detail.
Separate data
Training, tuning and test sets are kept apart. The final test set is to be held by the study statistician.
Locked models
A model version is frozen before it is evaluated, and any change triggers re-validation.
Representative cases
Datasets include the conditions that cause false positives, not only clear cancers and healthy controls.
Prespecified metrics
Performance measures and acceptance thresholds are written down before testing.
Bias checks
Results are examined across sites, instruments, lots and patient subgroups.
Transparent reporting
Studies are reported under the relevant guidelines, such as STARD for diagnostic accuracy studies and TRIPOD+AI for prediction models.
How it connects to the rest of the platform
The software is designed to take its input from the digital fluorescence reader and its calibration from the assay lot. Its outputs would feed the analytical and clinical studies in the clinical development plan. For the wider context, see our overview of AI in cancer diagnostics.
Partner roles in this work
We are looking for biostatisticians to own analysis plans and independent validation, data scientists with experience in signal or image analysis, and groups with expertise in regulatory requirements for software in medical devices. Academic partners can lead this as a distinct aim. See partner roles and agreements and the R&D priorities.
Frequently asked questions
Is the AI trained on patient data?
Not yet. Model development on clinical data will begin only with specimens and data collected under approved protocols.
Will the software make a diagnosis?
No. The research goal is decision support: a quantified result with quality information, interpreted by a clinician.
How would a learned model be regulated?
Software that is part of an in vitro diagnostic is reviewed with the device. Plans for any model changes after authorization would be discussed with regulators in advance.
References and further reading
- Collins GS, Moons KGM, Dhiman P, et al. TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods. BMJ. 2024;385:e078378.
- U.S. Food and Drug Administration, Health Canada, MHRA. Good Machine Learning Practice for Medical Device Development: Guiding Principles. 2021.
Lead the data science aim
A statistician or data science group can own this part of a proposal from design through publication.
Research and regulatory status
OncoFirm™ technologies described on this site are in research and development. Their performance characteristics have not been established, and no OncoFirm product has been cleared, approved or authorized by the U.S. Food and Drug Administration. Nothing here is medical advice. Reference to a federal agency or funding program does not imply endorsement, funding, sponsorship or affiliation.