MedicalRec: Medical recommender system for image classification without retraining

Researchers have developed MedicalRec, a transformer-based recommender system that identifies optimal deep learning models for medical image classification tasks without requiring retraining. The system leverages a new dataset (MedicalRec-Bench) containing over 5,000 model performance records across five medical imaging domains, achieving a 75.5% HitRate@100 and addressing the computational waste inherent in trial-and-error model selection.

The healthcare AI sector faces a significant efficiency problem: researchers typically select classification models through resource-intensive experimentation, consuming substantial computing power, energy, and generating electronic waste in the process. MedicalRec directly addresses this inefficiency by providing intelligent recommendations for which pre-existing models perform best on specific medical imaging tasks without requiring retraining or deployment of multiple candidates.

This research emerges from a broader industry recognition that model selection represents a hidden cost in AI development. Rather than inventing novel architectures, the study focuses on intelligent resource allocation—a pragmatic approach gaining traction as organizations prioritize sustainability and operational efficiency. The creation of MedicalRec-Bench, compiled from 3,000 published articles, represents a significant contribution to machine learning reproducibility, though the dataset contains substantial missing values due to inconsistent reporting standards in academic literature.

The system's 75.5% HitRate@100 performance across diverse medical domains—skin cancer, tumors, wounds, breast cancer, and MRI analysis—demonstrates meaningful predictive capability. For healthcare institutions and researchers, this translates to faster model deployment, reduced energy consumption, and lower environmental impact. The open-source availability through GitHub democratizes access to both the recommender system and benchmark dataset, potentially influencing how medical AI projects approach model selection going forward.

The next phase involves addressing dataset incompleteness and expanding the benchmark to additional medical imaging modalities. Widespread adoption depends on the research community's willingness to adopt standardized reporting practices and integrate recommendation systems into their development workflows.

• →MedicalRec uses transformer-based recommendations to eliminate trial-and-error model selection in medical imaging, reducing computational waste.
• →The publicly available MedicalRec-Bench dataset contains 5,000+ performance records across five medical imaging classification tasks.
• →The system achieved 75.5% HitRate@100, demonstrating strong performance across diverse medical domains without requiring model retraining.
• →Missing values in the dataset reflect broader reproducibility challenges in academic machine learning publishing standards.
• →Open-source release enables wider adoption and could establish standardized model selection practices in healthcare AI development.