PHAROS-AFE-AIMI: Multi-source & Fair Disease Diagnosis

Dimitrios Kollias, Anastasios Arsenos, Stefanos Kollias; Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) Workshops, 2025, pp. 7265-7273

The PHAROS-AFE-AIMI Workshop focuses on the application of trustworthy AI to medical imaging. It advances the AI-MIA Workshop series composed of 4 Workshops in 2021-2024, linking it to the Pharos AI Factory and specifically to its Healthcare vertical. Specific technologies are developed which target explainability, fairness, regularization, continual learning and domain shift analysis. Various medical imaging problems are tackled, including lung disease, cancer diagnosis, MRI, CT scan semantic segmentation and classification. Moreover, a competition was organized, comprising two tracks: (i) Multi-Source COVID-19 Detection Challenge, in which optimal systems were targeted that generalize across acquisition/site variations, (ii) Fair Disease Diagnosis Challenge, targeting CT scan classification to Healthy, Adenocarcinoma, Squamous Cell Carcinoma, or COVID-19, both in male and female categories. A baseline system has been developed employing a unified 3D convolutional encoder with sequence (RNN) aggregation for volumetric context, trained with standardized preprocessing and augmentation.