正確的處方配藥—藥袋中的排藥與處方箋的內容一致—是病人用藥安全最基本的前提。然而一旦涉及的藥物種類繁多即有人為失誤的可能。將有極高識別效果的深度學習技術，應用在輔助藥師進行即時性配藥核實的需求上理應值得期待，然而，在現實中，處方配藥台中的排藥種類繁多、包裝相似，並且可用於模型訓練的排藥影像樣本相對稀少，導致深度學習技術在排藥辨識應用上難以施展。本文提出「誘導式深度學習」之概念，透過人工的引導，讓深度學習系統能更加專注於重要特徵的學習，提升排藥辨識的效果。本文利用自建之成人錠劑排藥影像資料集，驗證了「誘導式深度學習」在各種經典的物體識別架構 (YOLO v2、ResNet、SENet) 都能產生顯著的識別效果提升；同時，本文也透過嵌入式技術，將排藥識別系統推向實時化、應用化的產品雛型階段，為進一步商品化打下了堅實的基礎。

Correct filling a prescription is of paramount importance to patient’s treatment and can be critical to parent’s life. Leveraging deep learning technique’s superior performance in identifying objects seems a promising solution to assist pharmacists in filling prescriptions. However, considering the huge number of blister package types involved, common appearance similarity present, and mostly importantly scarce data quantity encountered, direct application of deep learning results in less than desirable performance. This paper proposes a highlighted deep learning (HDL) approach to address the problem. Features are highlighted before entering the learning process so that most critical features can be learned with focus. Based on an Adult Lozenge dispensing station at MacKay Memorial Hospital, a pool of 250 types of blister packages, with 54 images of each type for training and 18 images for testing, has been collected. Three classic object identification networks of YOLO v2, ResNet, and SENet, have been used to implement the proposed HDL approach, all achieving close to the perfect overall performance. Meanwhile, the proposed solution has been implemented as a prototype in an embedded system that takes into account of requirements like real-time response and cost, making the proposed HDL solution halfway to commercial products.

[1] J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger, " 2016.
[2] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, and S. Reed, "SSD: Single Shot MultiBox Detector," arXiv preprint arXiv:1512.02325, 2015.
[3] K. He, X. Zhang, S. Ren, and J. Sun, "Deep residual learning for image recognition, " in Proc. IEEE conference on computer vision and pattern recognition, Las Vegas, United States, 2016, pp.770-778.
[4] J. Hu, L. Shen, and G. Sun, "Squeeze-and-excitation networks, " unpublished.
[5] M. A. Veronin, "Health Care Informatics: A Skills-Based Resource," American Journal of Pharmaceutical Education, vol. 71, no. 6, p. 126, 2007.
[6] C. A. Pedersen, P. J. Schneider, and D. J. Scheckelhoff, "ASHP national survey of pharmacy practice in hospital settings: dispensing and administration—2014," Am J Health Syst Pharm, vol. 72, pp. 1119-1137, 2015.
[7] E. G. Poon, J. L. Cina, W. Churchill, N. Patel, E. Featherstone, J. M. Rothschild, et al., "Medication dispensing errors and potential adverse drug events before and after implementing bar code technology in the pharmacy," Annals of internal medicine, vol. 145, pp. 426-434, 2006.
[8] J. A. Harolds and L. B. Harolds, "Quality and Safety in Health Care, Part X: Other Technology to Reduce Medication Errors," Clinical nuclear medicine, vol. 41, pp. 376-378, 2016.
[9] H. M. SEIDLING and D. W. BATES, "Evaluating the Impact of Health IT on Medication Safety," Evidence-Based Health Informatics: Promoting Safety and Efficiency Through Scientific Methods and Ethical Policy, vol. 222, p. 195, 2016.
[10] 林盈初, "即時性RFID智慧型藥櫃," 國立陽明大學醫學工程研究所論文, 2011.
[11] G. L. Cochran, R. S. Barrett, and S. D. Horn, "Comparison of medication safety systems in critical access hospitals: Combined analysis of two studies," American Journal of Health-System Pharmacy, vol. 73, pp. 1167-1173, 2016.
[12] Y. Lee, U. Park, A. Jain and S. Lee, "Pill-ID: Matching and retrieval of drug pill images," Pattern Recognition Letters，Vol. 33, no.7, pp. 904-910, May 2012.
[13] O. Taran, S. Rezaeifar, O. Dabrowski, J. Schlechten, T. Holotyak, and S. Voloshynovskiy, PharmaPack: Mobile fine-grained recognition of pharma packages. 2017, pp. 1917-1921.
[14] K. Saitoh, "Pharmaceutical Blister Pack Recognition using Local Features," in MVA, 2013, pp. 355-358.
[15] C. E. Pardo, L. F. Sosa, E. A. Gutierrez, and F. R. Jiménez, "Classification system for blister pack of pills," in 2014 IEEE 5th Colombian Workshop on Circuits and Systems (CWCAS), 2014, pp. 1-6.
[16] Y. Wong, H. Ng, K. Leung, et al. "Development of fine-grained pill identification algorithm using deep convolutional network," Journal of Biomedical Informatics，Vol. 74，pp. 130-136, Oct. 2017
[17] Y. Wang, J. Ribera, C. Liu, S. Yarlagadda, and F. Zhu, "Pill Recognition Using Minimal Labeled Data," in 2017 IEEE Third International Conference on Multimedia Big Data (BigMM), 2017, pp. 346-353.
[18] Szegedy, Christian, et al. "Going deeper with convolutions." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015.
[19] X. Zeng, K. Cao, and M. Zhang, "MobileDeepPill: A Small-Footprint Mobile Deep Learning System for Recognizing Unconstrained Pill Images," presented at the Proceedings of the 15th Annual International Conference on Mobile Systems, Applications, and Services, Niagara Falls, New York, USA, 2017.
[20] Y. Lecun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998.
[21] Krizhevsky, Alex, Ilya Sutskever, and Geoffrey E. Hinton. "Imagenet classification with deep convolutional neural networks." Advances in neural information processing systems. 2012.
[22] S. Karen, and A. Zisserman, "Very deep convolutional networks for large-scale image recognition," unpublished.
[23] Girshick, Ross, et al. "Rich feature hierarchies for accurate object detection and semantic segmentation." Proceedings of the IEEE conference on computer vision and pattern recognition. 2014.
[24] Girshick, Ross. "Fast r-cnn." Proceedings of the IEEE International Conference on Computer Vision. 2015.
[25] S. Ren, K. He, R. Girshick, et al. "Faster R-CNN: Towards real-time object detection with region proposal networks," in Proc. 29th Annual Conference on Neural Information Processing Systems, Montreal, Canada, 2015, pp. 91-99.
[26] J. Redmon, S. Divvala, R. Girshick, and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 779-788.
[27] F. Jiang, Q. Gu, H. Hao, et al. "Survey on Content-Based Image Segmentation Methods," Journal of Software, vol. 28,no. 1, pp. 160-183, Jan. 2017
[28] P. F. Felzenszwalb and D. P. Huttenlocher, "Efficient Graph-Based Image Segmentation," International Journal of Computer Vision, vol. 59, no. 2, pp. 167-181, 2004/09/01 2004.
[29] Y. Boykov and G. Funka-Lea, "Graph Cuts and Efficient N-D Image Segmentation," International Journal of Computer Vision, vol. 70, no. 2, pp. 109-131, 2006/11/01 2006.
[30] L. Zhengqin and C. Jiansheng, "Superpixel segmentation using Linear Spectral Clustering," in 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2015, pp. 1356-1363.
[31] A. Hartl, C. Arth, and D. Schmalstieg, "Instant segmentation and feature extraction for recognition of simple objects on mobile phones," in 2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition - Workshops, 2010, pp. 17-24.
[32] D. Powers and Ailab, Evaluation: From precision, recall and F-measure to ROC, informedness, markedness & correlation. 2011, pp. 2229-3981.