How we train AI models to detect tumors and genetic mutations
Researchers trained artificial intelligence models using the TCGA BRCA and LUSC datasets for the detection of tumor tissue and the TP53 mutation. Work was carried out on histological section images in different formats and magnifications, comparing paraffin-fixed processed slides with frozen ones and evaluating multiple magnification levels to identify the most informative scales for each task.
Methodology and key observations: the complete images were preprocessed through segmentation and fragmentation into patches to feed models based on convolutional neural networks and transfer learning approaches. The results showed variations in accuracy depending on the type of section and magnification: FFPE slides improved TP53 mutation detection, probably due to better morphological preservation and molecular signal; meanwhile, frozen slides were sufficient for robust tumor detection in many configurations, especially at intermediate magnifications. Overall, low magnifications captured global tissue architecture and high magnifications allowed detecting cellular details relevant for molecular predictions.
Validation and generalization: to ensure generalization capability, cross-validation and optimal sampling strategies were applied, in addition to balancing classes and controlling sources of bias between cohorts. These techniques helped stabilize performance metrics and identify training configurations that generalize better to external sets. In summary, the combination of proper selection of tissue type, magnification, and sampling techniques improved the detection of both tumors and specific mutations such as TP53.
About Q2BSTUDIO: at Q2BSTUDIO we are a custom software and application development company specialized in solutions based on artificial intelligence and cybersecurity. We offer custom software services and custom applications that integrate AI models to solve real challenges in the healthcare, industry, and business sectors. Our artificial intelligence specialists design scalable pipelines, while the cybersecurity team ensures the protection of sensitive data. We also provide aws and azure cloud services and business intelligence services so organizations can deploy and monitor solutions with full confidence.
Featured services: we develop custom software, custom applications, artificial intelligence solutions and AI for businesses, custom AI agents, and power bi dashboards for advanced analytics. Our catalog includes integration with aws and azure cloud services, implementation of business intelligence services, and cybersecurity strategies tailored to each client.
Contact and value proposition: if your organization seeks to carry out automatic tumor detection projects, genetic mutation analysis, or implement artificial intelligence at scale, Q2BSTUDIO accompanies from consulting to production deployment. We combine technical expertise in artificial intelligence, custom software development, AI agents, power bi, and aws and azure cloud services with cybersecurity practices to offer secure, efficient, and results-oriented solutions.
Keywords: custom applications, custom software, artificial intelligence, cybersecurity, aws and azure cloud services, business intelligence services, artificial intelligence, AI for businesses, AI agents and power bi.
