Abstract
There has been much progress in data-driven artificial intelligence technology for medical image analysis in last decades. However, it still remains a challenge due to its distinctive complexity of acquiring and annotating image data, extracting medical domain knowledge, and explaining the diagnostic decision for medical image analysis. In this paper, we propose a data-knowledge-driven evolutionary framework termed as Parallel Medical Imaging (PMI) for medical image analysis based on the methodology of interactive ACP-based parallel intelligence. In the PMI framework, computational experiments with predictive learning in a data-driven way are conducted to extract medical knowledge for diagnostic decision support. Artificial imaging systems are introduced to select and prescriptively generate medical image data in a knowledge-driven way to utilize medical domain knowledge. Through the parallel evolutionary optimization, our proposed PMI framework can boost the generalization ability and alleviate the limitation of medical interpretation for diagnostic decision.
Introduction
Despite the rapid advancement of deep learning in computer vision, medical image analysis faces unique hurdles that prevent widespread clinical adoption. The primary challenges include Data Scarcity, where high-quality annotated images are difficult to obtain due to privacy and expert time constraints; the Knowledge Gap, where models often ignore established "top-down" medical expertise in favor of raw pixel patterns; and Interpretability, the "black box" nature of AI that fails to provide the clinical reasoning necessary for medical decisions.
To address these issues, we propose the Parallel Medical Imaging (PMI) framework. PMI bridges these gaps by treating medical diagnosis as a parallel interaction between the physical world (real patients) and an artificial world (simulated data and models), ensuring that AI systems are both data-driven and knowledge-guided.
The PMI Framework and ACP Methodology
PMI is built upon the ACP methodology: Artificial systems, Computational experiments, and Parallel execution. This framework creates a closed-loop system where data and knowledge evolve together.
1. Artificial Imaging Systems (AIS)
Instead of relying solely on limited real-world datasets, PMI utilizes AIS to generate "Artificial Big Data." Using generative models like GANs and computer graphics, the system creates synthetic images. Crucially, this generation is prescriptive: it uses medical rules to generate specific lesion types (e.g., spiculated margins) to explicitly teach the model critical diagnostic features.
2. Computational Experiments
The model undergoes Predictive Learning on a hybrid dataset of real and synthetic images. This allows the AI to extract "Specific Small Knowledge"—the key diagnostic features that align with clinical reality.
3. Parallel Execution
This stage involves a continuous optimization loop. The diagnostic model's failures inform the AIS on what kind of synthetic data to generate next (Descriptive Learning), while medical domain knowledge guides that generation (Prescriptive Learning).
Case Study: Mammogram Analysis
To demonstrate the efficacy of PMI, we applied it to mammogram mass detection and classification using the INbreast dataset. A key technique used was "synthesis by insertion," where realistic masses are generated and placed into normal mammograms to expand the training set.
In our experiments, a standard CNN achieved an Accuracy of 85% and an AUC of 0.859. By integrating the PMI framework (incorporating predictive, descriptive, and prescriptive learning), the Accuracy rose to 90.2% and the AUC to 0.901.
Case Study: Skin Lesion Analysis
We further validated PMI on the ISIC Skin 2017 dataset. By combining modern deep learning with "hand-crafted" features (mathematical descriptions of texture and color), the PMI approach achieved an AUC of 0.912. This significantly outperformed the baseline model (AUC 0.667), demonstrating the framework's robustness when real-world data is scarce.
Conclusion
The Parallel Medical Imaging framework represents a shift from purely data-driven AI to a hybrid data-knowledge-driven approach. By utilizing artificial systems to bridge the data gap and prescriptive learning to incorporate domain expertise, PMI enhances both the performance and the interpretability of medical AI. Future work will focus on scaling this framework to 3D imaging modalities like CT and MRI and integrating multi-modal patient data into the parallel execution loop.
Summary updated