Abstract
There has been much progress in data-driven artificial intelligence technology for medical image analysis in the last decades. However, it still remains challenging 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 framework termed as Parallel Medical Imaging (PMI) for intelligent 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 closed-loop optimization based on parallel execution, our proposed PMI framework can boost the generalization ability and alleviate the limitation of medical interpretation for diagnostic decisions. Furthermore, we illustrate the preliminary implementation of PMI method through the case studies of mammogram analysis and skin lesion image analysis. Experimental results on several public medical image datasets demonstrate the effectiveness of proposed PMI.
Introduction: Bridging the Gap Between Data and Doctors
Artificial Intelligence has revolutionized medical image analysis, but it faces a persistent "triple threat" of challenges: the scarcity of high-quality annotated data, the difficulty of integrating centuries of human medical expertise, and the "black box" nature of deep learning models. To address these hurdles, researchers have proposed a new framework called Parallel Medical Imaging (PMI).
PMI shifts the paradigm from purely data-driven learning to a hybrid system that utilizes "Parallel Intelligence." By creating a closed-loop system where artificial imaging systems generate synthetic data guided by medical knowledge, PMI improves diagnostic accuracy and makes AI decisions more interpretable for clinicians.
The PMI Framework: The ACP Methodology
The core of PMI is the ACP methodology—Artificial societies, Computational experiments, and Parallel execution—adapted for the medical imaging field.
- Artificial Imaging Systems (AIS): Instead of relying solely on real images, the system uses "Descriptive Learning" to build artificial environments. It can use Generative Adversarial Networks (GANs) or simulation techniques to create synthetic but medically realistic images (e.g., inserting a "spiculated" mass into a healthy mammogram) based on expert clinical descriptions.
- Computational Experiments: The system performs "Predictive Learning" on a mix of real and synthetic data. This allows the model to "experiment" with rare cases that aren't well-represented in real-world datasets, effectively turning "small data" into "big intelligence."
- Parallel Execution: This is the "closed loop" optimization. The system uses "Prescriptive Learning" to guide the generation of new data based on what the model struggles with. If a model fails to identify specific irregular lesions, the AIS is prescribed to generate more of those specific cases to improve performance.
Case Studies and Implementation
The researchers validated the PMI framework on two primary benchmarks: mammogram analysis (using the INbreast dataset) and skin lesion classification (using the ISIC Skin 2017 dataset).
Mammogram Analysis
In mammography, the framework uses a "knowledge-to-data" approach. By using medical descriptions (like "irregular shape" or "spiculated margin") to generate synthetic images, the framework explicitly bakes human expertise into the training data.
Skin Lesion Analysis
For skin lesions, the authors combined deep learning features with "hand-crafted" features (like HOG and SIFT) that mimic how dermatologists describe lesions (color, texture, symmetry). This hybrid feature embedding makes the model's internal logic more aligned with human reasoning.
Key Results
The PMI framework consistently outperformed standard deep learning models across both tasks:
- Mammogram Analysis: Accuracy rose from a baseline of 85% to 90.2%, with an Area Under the Curve (AUC) of 0.901.
- Skin Lesion Analysis: In melanoma detection, the PMI approach achieved an AUC of 0.912, significantly outperforming standard relation networks and top-ranking results from the ISIC 2017 challenge.
Conclusion and Future Directions
PMI demonstrates that high-level intelligence does not necessarily require millions of real images if one can intelligently generate data from a small set of high-quality examples. While currently focused on specific lesions, future work aims to expand PMI into complex, multi-organ diseases and clinical integration, where "virtual patients" can simulate treatments before they are applied in the real world.
Reference: C. Gou et al., "Parallel Medical Imaging for Intelligent Medical Image Analysis: Concepts, Methods, and Applications," IEEE/CAA Journal of Automatica Sinica, 2021.
Summary updated