ZiuZ Challenges
Pharmacies | 16.11.18 | 分钟阅读时间
Pill Challenge

过期

过期

Update: On Monday June 3rd, Ruth Wijma (24 years old, Master Student Artificial Intelligence at the University of Amsterdam) was proclaimed the winner of the ZiuZ Pill Challenge. The results of the algorithm developed by Ruth corresponded most with the predetermined ground truth. Interested? Please read our news item on the Pill Challenge.

ZiuZ offers the best machines for the quality control of the packaging of medicine bags or pouches. Saving pharmacists valuable time. Our checking devices automatically check the content of each bag, ensuring that they contain the right medication. The device photographs each bag and stores the picture for later reference. ZiuZ’s checking devices are compatible with a wide range of packaging robots. The machines analyses full color and contour images (where a backlight is used to shine through the bag). Our algorithm is able to detect and recognize medication by color, shape and size and is robust to several issues such as different font text, reflections, transparent medication, shades and presence of printed objects such as 2D barcodes. For ZiuZ, the patient safety comes first and that is why we will always look for new ways to further improve our algorithms.

The medication pouches and the medication itself are continuously changing, leading to new challenges. In order to keep offering innovation and quality, we are proposing the “Pill Challenge” event. This project consists of developing an automatic system which is able to accurately segment several kinds of meditations in pouches. Specifically, given a large collection of images, we want to answer the following questions in an objective, measurable way:

  • What are the best medication segmentation methods for transparent plastic pouches?
  • What is the fastest method to do so?

Doing the above represents one important step towards keeping automation great. Indeed, rather than having to manually browse and/or watch a collection of images, the system can basically automatically segment and count medication.

 

Database

We provide you with a training database of 1743 images (581 medication pouch) with a ground truth, and a testing database of 642 images (321 medication pouch) without the ground truth segmentation. Both databases contain different kinds of medications and pouches. The training database will be made available after completion of the registration and after you have accepted the terms and conditions of the event. The testing database will be made available on March the 15th, two weeks before the submission deadline. As input data for testing and validation, we provide a collection of images of several kinds of medications in pouches. For each medication pouch there are 2 images (RGB and contour image) as shown figure 1.

Figure 1. Medication pouch: (left) RGB colour image, (right) contour/backlit image.

The database covers a wide range of situations: different kinds of pouches, pouches with different amounts of medication and different types of medications (differing in size, shape and transparency).

 

Expected Results

There is not an identical work to this project in the literature, to our knowledge. However, there are numerous works about object segmentation on IEEE Xplore Digital Library, ScienceDirect and ACM Digital Library. Any method showing good results is welcome in this challenge. An ideal result for this project will be an algorithm able to:

  • Accurately segment all kinds of medication, including transparent;
  • Segment smaller contaminations (crumbs) which might be present in the bag;
  • Ignore any printing which is present on the bag;
  • Do that faster than half a second.

Implementation: Any method could be implemented and proposed to solve this challenge. Any programming language is acceptable as long as an executable can output a black\white image when given the two input images.

Final Results: The final results are binary images, black for background and white for segmented medication. We (ZiuZ) are going to qualitatively evaluate the final results (binary images) by matching with our ground truth results. The team with the best results will be invited to present their method at ZiuZ. We offer a prize of €2.500 to the winner. The participant should deliver for the final presentation:

  1. Source code;
  2. Documentation, including a detailed explanation of the algorithm and how to install, run and use the algorithm/software.

 

Important dates
  • Training database available: November 16th 2018
  • Testing database available: March 15th 2019
  • Submission of testing database results: April 15th 2019

The results of the testing database are binary images, black for background and white for segmented medication

  • Participants Notification: May 1st 2019
  • Final Presentation and prize: May 30th 2019

For the final presentation, we expect you to provide us with the executable/algorithm and a detailed description of how to install and use the solution.

 

Terms and conditions apply to this contest.

Interested?

Join the challenge