Histopathologic image analysis is the golden standard for diagnosis of malignant lesions, but manual examination of images causes intense workload for pathologists. Computer-aided diagnosis (CAD) systems shall improve the diagnosis efficiency and increase inter-observer agreement. The TissueGnostics GmbH develop deep learning-based CAD systems to support physicians and improve diagnosis for patients.

Eingereicht von: Rupert Ecker, PhD
Amirreza Mahbod
Robert Nica
Catalin Captarencu
Firma/Universität: TissueGnostics GmbH
Homepage: www.tissuegnostics.com
Kooperationspartner: Medical University Vienna

Histopathological image analysis is considered as the most reliable method for recognition and diagnosis of tissue abnormalities such as malignant lesions. Image analysis is usually performed by an experienced pathologist, but the diagnosis process is not trivial. Many studies have shown that the average diagnostic concordance between specialists is approximately 75%. The manual examination of histological images poses an intense workload on the pathologists; in addition, there is a lack of pathologists in many parts of the world. In summary, these limitations motivate the development of computer-aided diagnosis (CAD) systems.

CAD systems are semi or fully automated image analysis algorithms, which are developed to help pathologists during the diagnosis procedure. Being a second opinion system, CAD systems are supposed to reduce the workload of specialists, to improve the diagnosis efficiency, to increase the level of inter-observer agreement and, in the end, also contribute to cost reduction.

life-science.eu - Foto: (c) Rupert EckerWhile there are a large number of automatic or semi-automatic methods for image analysis, machine learning-based algorithms have shown to be superior over conventional image processing techniques when multiple pattern types have to be recognized and distinguished. Conventional machine learning algorithms are based on hand-crafted features which are derived from images and attempt to replicate the pathologist’s classification method. However, a major problem of these approaches is the difficulty in choosing the most meaningful features for a specific dataset at hand, which potentially degrades the performance of the algorithm and increases the model complexity to be used by non-experts.

The recent increase in available computing power and training dataset sizes allowed for the application of Convolutional Neural Networks (CNNs) to image classification and segmentation problems. In contrast to conventional machine learning approaches for image classification, they do not rely on hand-crafted features, but utilize large amount of images to derive task-specific image features. So far, these deep learning models have achieved excellent performance in image classification challenges in different fields, including medical image analysis, and in particular on histopathological images.

CAD may yield relevant information for diagnosis and prognosis, which the human eye and mind are unable to recognize or appreciate and thereby help to find yet unknown subclasses of histological features and/or diseases. While extremely promising, it will still take a significant amount of research and validation before the patient and physicians will benefit from this type of application. Thus, the research community still has to develop – and validate – appropriate algorithms.

The long-lasting collaboration between the company TissueGnostics GmbH (headed by CEO Rupert Ecker), an expert in microscope automation and tissue analysis, and the Medical University of Vienna (Isabella Ellinger, PI at the Inst. f. Pathophysiology and Allergy Research) is currently fostered by the joint participation in an EU-funded European Training Network (ETN; Grant #675228; https://casr.meduniwien.ac.at). The common PhD student, Amirreza Mahbod, makes use of deep CNNs for segmentation and classification of different histopathological images with minimum pre- and post-processing steps. In order to implement these algorithms, state-of-the-art deep learning-based architectures pre-trained on huge data sets of natural images are utilized and modified to be adapted for histopathological image analysis. All algorithms shall be tested and validated in „grand challenges in biomedical image analysis“. Upon successful performance of algorithms, the ultimate goal is to integrate them into the commercial StrataQuestTM software of TissueGnostics GmbH (Catalin Captarencu), make them available for clinical as well as biomedical research applications and integrate them into the custom StrataQuest™ Apps provided by TissueGnostics (Robert Nica).