The Impact of AI in Healthcare and the Challenge of Consumption

The emergence of machine learning and deep learning has revolutionized the efficiency of diagnostic, therapeutic, and administrative systems in healthcare. However, this rapid adoption has come at the cost of requiring significant computing power, high energy consumption, e-waste generation, and an increasing carbon footprint. For organizations evaluating on-premise deployments, these factors directly translate into higher Total Cost of Ownership (TCO) and sustainability challenges.

One of the main difficulties in implementing these models lies in choosing the right model for specific classification tasks. Traditionally, researchers attempt to identify the optimal model for their data through a trial-and-error process. This iterative approach is inherently inefficient, as it involves considerable energy and computational resource expenditure, generating waste that can be avoided.

MedicalRec: A Transformer-Based Approach to Optimization

To address these issues, MedicalRec, a model-based recommender system, has been developed. The primary goal of this study is to provide a system that guides the selection of the most appropriate model for medical image classification, eliminating the need for extensive retraining or re-evaluation cycles. MedicalRec is based on a transformer architecture, a widely adopted solution for item recommendation tasks.

For its creation, a vast dataset was collected from 3,000 scientific articles in the field of medical image classification. This dataset, publicly available under the name MedicalRec-Bench, contains over 5,000 records of models tested in various applications, including Skin Cancer Classification, Tumour Classification, Wound Classification, Breast Cancer, and MRI classification. The dataset was evaluated in four different configurations, depending on the number of features considered: MedicalRec I (5 features), MedicalRec II (9 features), MedicalRec III (11 features), and MedicalRec IV (18 features). Despite its comprehensiveness, collecting all values for the features presents challenges due to non-reporting by the original authors, leading to significant amounts of missing values in the dataset. In evaluations conducted on the dataset and with 12 base models, MedicalRec achieved remarkable results, reaching a maximum HitRate@100 of 75.5%. The dataset and implementations are available via a public GitHub link.

Implications for On-Premise Deployments and Data Sovereignty

Optimizing model selection, as proposed by MedicalRec, has direct and significant implications for organizations opting for on-premise deployments or air-gapped environments. Reducing the trial-and-error phase means minimizing the consumption of hardware resources – such as GPUs and VRAM – and the energy required for training and inference. This translates into lower TCO, greater operational sustainability, and a reduced environmental footprint, all crucial factors for CTOs and infrastructure architects.

In healthcare contexts, data sovereignty and regulatory compliance (such as GDPR) are absolute priorities. A system that allows for rapid identification of efficient models without needing to move sensitive data to external cloud platforms strengthens organizations' ability to maintain complete control over their information assets. While MedicalRec is not an LLM, its optimization logic aligns perfectly with the AI-RADAR philosophy, which promotes solutions that improve efficiency and control in self-hosted AI workloads. For those evaluating on-premise deployments, analytical frameworks are available on /llm-onpremise to assess specific trade-offs related to hardware, costs, and compliance requirements.

Future Prospects and the Importance of Efficiency in Medical AI

The introduction of systems like MedicalRec marks a step forward towards a more conscious and efficient adoption of artificial intelligence in critical sectors such as medicine. The ability to select optimal models with less resource waste not only accelerates the development and deployment of AI solutions but also contributes to mitigating the environmental impact of technological infrastructure.

Looking ahead, the expansion of approaches similar to MedicalRec could extend to other AI domains, fostering a culture of efficiency and sustainability. For technical decision-makers, investing in tools that optimize computational resource usage represents a key strategy for balancing innovation, operational costs, and environmental responsibility, especially in a landscape where AI workloads continue to grow in complexity and hardware requirements.