Image analysis is done on a daily basis in basic research as well as in a clinical setting. A variety of tools with differing advantages and limitations are available to analyze images. Amongst these, there are tools like ImageJ/Fiji (open source) (Schindelin et al. 2012; Schneider et al. 2012) that are often used in basic research and comprise powerful tools to create complex image analysis pipelines. Medical images including x-rays, CTs, or MRIs are usually visualized on PACS-viewers with basic tools for measurements, while more complex measurements (e.g. joint replacement planning) are performed by exporting the images to tools like mediCAD® (mediCAD Hectec GmbH, Altdorf, Germany). The above-mentioned tools are mostly desktop-bound and results are stored separately.
Survey-tools like SurveyMonkey® (SurveyMonkey Inc., San Mateo, California/USA) or Qualtrics® (Qualtrics International Inc., Provo, Utah/USA) can be used to reach out to multiple observers but are more qualified to gather opinions rather than analyze data. Besides, they offer limited possibilities to display medical images or tools for measurements. Each question and answer is created individually which can be challenging in projects comprising many images to analyze where a project master would have to create the same question n-times, one after another.
Another way to analyze images is by employing artificial intelligence (Pedoia and Majumdar 2018; Pedoia et al. 2018; Lee et al. 2020). This highly efficient method relies largely on algorithms that can process vast amounts of data without the risk of human error. Yet the accuracy of these algorithms is not only based on their code but also on the quality of the images, they were trained with. Additionally, setting up an algorithm for specific analyses can be laborious and require some technical skills.
The goal of TYCHE was to create a lightweight browser-based tool to simplify image analysis by multiple observers, even across labs and countries. It certainly has – as of today – rather basic measurement tools in contrast to advanced desktop-bound tools. However, results are stored online and can immediately be seen by the project master. In contrast to existing survey tools, it is well prepared for scenarios in which the same score or measurement has to be applied on multiple images. Questions/tasks and answers are only created once and are then applied to all the uploaded images within one project. In case a different question/task is asked for each image, a classic survey tool is the better option.
Blinding and randomization during data analysis is gaining more and more importance in science to increase objectivity. Nowadays, journals such as Nature request information about blinding and randomization in studies, including if and how randomization was achieved, if data analysis was done blinded and, if not, why randomization/blinding were not done. (https://www.nature.com/documents/nr-reporting-summary-flat.pdf, accessed on Nov 4th, 2021). Sometimes, group allocation is disclosed due to sample labelling, and especially when more than one staining is applied to tissue sections, bias can be introduced in a way that features are over- or underrated, dependent on group allocations. Furthermore, if randomization was not possible during sample preparation, specimens of one treatment group might come after each other and are analyzed in that same order. Here, Tyche provides an easily applicable solution, as pictures are displayed without annotations/file names visible to the observer and in a randomized order. The same holds true for clinical data that is analyzed for a specific scientific context, also here, blinding for conditions and randomization is highly advised. It has to be noted, that Tyche does not anonymize the image files itself but displays them anonymized. It is impossible for the observer to view the filename or any metadata stored in the header of the image. Neither the image nor the image storage path can be accessed directly as opposed to images normally displayed on websites where all this information can be visualized with a right-click. Meta-data or any information that had been written or painted on the image using a graphic program like Microsoft Paint® or Photoshop® can not be removed by Tyche.
An increasing number of scores and grading systems is being established in basic research and medicine. The OARSI-score, which was used in this study, can be considered a broadly accepted means of analyzing histological images in the scientific community working on osteoarthritis. In the clinical environment, scores and grades applied to medical images are a substantial part in finding the right treatment options and thus the demand for reliable analysis is high. With TYCHE we aim to facilitate and standardize the approach of analyzing images while trying to minimize subjectivity and bias. In the present study, we used the OARSI-score to validate TYCHE for the use of a score to analyze images. As opposed to the traditional offline analysis on a desktop, images were analyzed in a random order by multiple blinded observers online. Results were immediately visible for the project master.
In summary, we present a novel, free, browser-based tool to facilitate image and video analysis by multiple observers. By displaying anonymized data in a random order, TYCHE helps to reduce subjectivity and bias. Furthermore, as the analysis is done online, file sharing of the imaging material with observers as well as gathering results via spreadsheets become obsolete, and therefore Tyche facilitates collaborations of different labs.
A free TYCHE-account can be requested at [email protected].