More than 300 million people around the world suffer from some rare disease, but their individual diagnosis is still a challenge. The use of Artificial Intelligence (AI) techniques is significantly changing the analysis of biomedical images, accelerating and improving the diagnosis process. However, training an effective AI model often necessitates a large dataset, particularly for the most advanced methodologies. As a result, not all medical specialties benefit from AI techniques to the same extent. For instance, AI techniques’ application in diagnosing rare diseases is limited, as the low prevalence and high complexity of these conditions lead to insufficient data for effectively training AI models. In this thesis, we will address these limitations by designing AI algorithms optimized for small and complex databases. The Sant Joan de Déu Hospital is an international reference centre for rare diseases research, specialized in neurogenetic and paediatric oncology rare diseases. Usually, genetic tests are not enough to completely assess these diseases. Therefore, the visual inspection of the cellular state is crucial for gaining meaningful insights into the disease’s pathophysiology. Novel advanced optical microscopy techniques have the potential to significantly improve these procedures. However, they are still under development and their integration with AI methodologies has barely been explored. This thesis will combine advanced optical microscopy technologies, complex image analysis pipelines and new AI techniques to train models robust for use with scarce databases. This work will be developed under the guidance of Universitat Pompeu Fabra and Hospital Sant Joan de Deu’s professionals. We hope to transform the diagnosis scope of rare disease through this work. The main efforts will focus on overcoming the challenge of using small and complex databases. For this purpose, novel AI algorithms, new training techniques, and synthetic data generation technologies will be implemented. In the end, all these novel strategies will give new insights into the challenges of rare diseases assessment and treatment monitoring, so that they can take advantage of the revolution that the biomedical image field is undergoing in the same way as many other biomedical areas.More than 300 million people around the world suffer from some rare disease, but their individual diagnosis is still a challenge. The use of Artificial Intelligence (AI) techniques is significantly changing the analysis of biomedical images, accelerating and improving the diagnosis process. However, training an effective AI model often necessitates a large dataset, particularly for the most advanced methodologies. As a result, not all medical specialties benefit from AI techniques to the same extent. For instance, AI techniques’ application in diagnosing rare diseases is limited, as the low prevalence and high complexity of these conditions lead to insufficient data for effectively training AI models. In this thesis, we will address these limitations by designing AI algorithms optimized for small and complex databases. The Sant Joan de Déu Hospital is an international reference centre for rare diseases research, specialized in neurogenetic and paediatric oncology rare diseases. Usually, genetic tests are not enough to completely assess these diseases. Therefore, the visual inspection of the cellular state is crucial for gaining meaningful insights into the disease’s pathophysiology. Novel advanced optical microscopy techniques have the potential to significantly improve these procedures. However, they are still under development and their integration with AI methodologies has barely been explored. This thesis will combine advanced optical microscopy technologies, complex image analysis pipelines and new AI techniques to train models robust for use with scarce databases. This work will be developed under the guidance of Universitat Pompeu Fabra and Hospital Sant Joan de Deu’s professionals. We hope to transform the diagnosis scope of rare disease through this work. The main efforts will focus on overcoming the challenge of using small and complex databases. For this purpose, novel AI algorithms, new training techniques, and synthetic data generation technologies will be implemented. In the end, all these novel strategies will give new insights into the challenges of rare diseases assessment and treatment monitoring, so that they can take advantage of the revolution that the biomedical image field is undergoing in the same way as many other biomedical areas.