Adaptive prompt engineering for few-shot medical image segmentation via meta-learning: A framework with visual selection, knowledge enhancement, and volumetric refinement

Dalia N. Abdul-Wadood

Authors

Dalia N. Abdul-Wadood University of Baghdad, Iraq

Keywords:

Few-shot segmentation, Medical image analysis, Meta-learning, Prompt engineering

Abstract

Medical image segmentation typically requires large-scale pixel-level annotations, limiting clinical applicability. the proposed meta-learning framework for few-shot segmentation that adaptively engineers visual and textual prompts without manual intervention. The framework introduces three key contributions: (i) a Meta-driven Visual Prompt Selection (MVPS) mechanism that retrieves most of the informative support images via episodic reinforcement learning; (ii) a Knowledge-Enhanced Prompt-Image Learning (KEPIL) module that integrates structured anatomical ontologies (Uberon, GPT-4 generated descriptors); and (iii) an Inference-Time Pseudo-Labeling (ITPL) strategy that exploits unlabeled volumetric slices. Evaluated on six public datasets (BTCV, CHAOS, LiTS, KiTS, DRIVE, FLARE2022), our method achieves a mean Dice of 88.7% (averaged across all datasets) with 5 support images, outperforming state-of-the-art methods with an average margin of 4.3% (p < 0.01 after Bonferroni correction).

References

[1] M. Li, Y. Jiang, Y. Zhang, and H. Zhu, “Medical image analysis using deep learning algorithms,” Front. Public Health, vol. 11, Art. no. 1273253, 2023.

[2] M. I. Razzak, S. Naz, and A. Zaib, “Deep Learning for Medical Image Processing: Overview, Challenges and the Future,” in Classification in BioApps: Automation of Decision Making, Cham, Switzerland: Springer, 2017, pp. 323–350.

[3] A. Tiwari, S. Mishra, and T.-R. Kuo, “Current AI Technologies in Cancer Diagnostics and Treatment,” Mol. Cancer, vol. 24, no. 1, Art. no. 159, 2025.

[4] J. Banerjee, J. N. Taroni, R. J. Allaway, D. V. Prasad, J. Guinney, and C. Greene, “Machine Learning in Rare Disease,” Nat. Methods, vol. 20, no. 6, pp. 803–814, 2023.

[5] Y. Gao, Y. Jiang, Y. Peng, F. Yuan, X. Zhang, and J. Wang, “Medical Image Segmentation: A Comprehensive Review of Deep Learning-Based Methods,” Tomography, vol. 11, no. 5, Art. no. 52, 2025.

[6] A. Ray, F. Firouzi, B. Farahani, and K. Chakrabarty, “Computer Vision for Healthcare and Medicine: Unlocking Insights from Visual Data,” in Smart and Connected Health: AI, IoT, and Trustworthy Technologies, Cham, Switzerland: Springer Nature Switzerland, 2026, pp. 423–472.

[7] C. Wu, D. Restrepo, Z. Shuai, Z. Liu, and L. Shen, “Efficient In-Context Medical Segmentation with Meta-Driven Visual Prompt Selection,” in Proc. Int. Conf. Medical Image Computing and Computer-Assisted Intervention (MICCAI), Cham, Switzerland: Springer Nature Switzerland, 2024, pp. 255–265, doi: 10.1007/978-3-031-72114-4_25.

[8] M. Ali, T. Wu, H. Hu, Q. Luo, D. Xu, W. Zheng, N. Jin, C. Yang, and J. Yao, “A Review of the Segment Anything Model (SAM) for Medical Image Analysis: Accomplishments and Perspectives,” Comput. Med. Imag. Graph., vol. 119, Art. no. 102473, 2025, doi: 10.1016/j.compmedimag.2024.102473.

[9] C. Chang, H. Law, C. Poon, S. Yen, K. Lall, A. Jamshidi, V. Malis, D. Hwang, and W. C. Bae, “Segment Anything Model (SAM) and Medical SAM (MedSAM) for Lumbar Spine MRI,” Sensors, vol. 25, no. 12, Art. no. 3596, 2025, doi: 10.3390/s25123596.

[10] J. Wang, Y.-I. Lin, and S.-Y. Hou, “A Data Mining Approach for Training Evaluation in Simulation-Based Training,” Comput. Ind. Eng., vol. 80, pp. 171–180, 2015, doi: 10.1016/j.cie.2014.12.008.

[11] R.-D. Wu, Y.-Y. Lin, and H.-F. Yang, “SQUARE: Semantic Query-Augmented Fusion and Efficient Batch Reranking for Training-Free Zero-Shot Composed Image Retrieval,” arXiv, arXiv:2509.26330, 2025, doi: 10.48550/arXiv.2509.26330.

[12] T. Dissanayake, Y. George, D. Mahapatra, S. Sridharan, C. Fookes, and Z. Ge, “Few-Shot Learning for Medical Image Segmentation: A Review and Comparative Study,” ACM Comput. Surv., vol. 58, no. 1, pp. 1–36, 2025, doi: 10.1145/3746224.

[13] W. Zhang, L. Luo, M. He, J. Hai, and J. Ye, “Descriptor MedSAM: Language-image fusion with multi-aspect text guidance for medical image segmentation,” Sci. Rep., vol. 16, Art. no. 3758, 2026, doi: 10.1038/s41598-025-33843-5.

[14] M. Lan, L. Zhang, and X. Li, “OFL-SAM2: Prompt SAM2 with Online Few-Shot Learner for Efficient Medical Image Segmentation,” in Proc. AAAI Conf. Artif. Intell., vol. 40, no. 7, pp. 5809–5817, Mar. 2026, doi: 10.1609/aaai.v40i7.37502.

[15] Y. Xie, S. Zhang, H. Cheng, P. Liu, Z. Gero, C. Wong, T. Naumann, H. Poon, and C. Rose, “DocLens: Multi-Aspect Fine-Grained Medical Text Evaluation,” in Proc. 62nd Annu. Meeting Assoc. Comput. Linguistics (ACL 2024), Vol. 1: Long Papers, Bangkok, Thailand, Aug. 2024, pp. 649–679.