Medical Biopharmaceutical Image Anomaly Detection under Retinex State Space Duality and Frequency Consensus-Driven Transformer

Yuhan Dai — Thu, 02 Apr 2026 00:00:00 +0800

With the rapid development of medical imaging technologies and biopharmaceutical research, a large amount of high-dimensional and complex medical and biological image data is continuously being generated. Achieving high-precision anomaly detection under conditions of complex backgrounds and low contrast has become an important research problem in the fields of intelligent healthcare and drug development. Traditional anomaly detection methods often struggle to achieve stable and robust detection performance when dealing with issues commonly present in medical images, such as uneven illumination, complex structures, and insufficient utilization of frequency information. To address these challenges, this paper proposes a Transformer-based anomaly detection method for medical and biopharmaceutical images driven by Retinex state-space duality and frequency consensus. By integrating spatial-domain and frequency-domain features, the proposed method enhances the model’s ability to perceive complex abnormal structures. Specifically, the method first employs Retinex state-space duality to decompose medical images into structural and illumination components, thereby strengthening the structural representation of anomalous regions while reducing interference caused by illumination variations. Subsequently, a frequency consensus-driven mechanism is introduced to model feature consistency across different scales in the frequency domain, enabling adaptive enhancement of key anomalous frequency features. On this basis, a global context modeling framework is constructed by incorporating a Vision Transformer, which captures long-range dependencies and potential anomalous patterns in medical images, further improving detection accuracy and feature representation capability. To verify the effectiveness of the proposed method, extensive experiments are conducted on multiple medical and biopharmaceutical image datasets, and comparisons are made with several mainstream anomaly detection models. Experimental results demonstrate that the proposed method outperforms the comparison methods across evaluation metrics, achieving more stable and accurate anomaly detection in complex medical imaging environments. This approach provides a technically promising solution with potential application value for automated image analysis in intelligent pharmaceutical inspection and drug development.

Journal of Computational Methods in Engineering Applications

Medical Biopharmaceutical Image Anomaly Detection under Retinex State Space Duality and Frequency Consensus-Driven Transformer