Image analysis and computational pathology

We use established image analysis softwares to improve biomarker assessment in routine pathology. We also develop in-house deep-learning based models for next generation image analysis together with the Rantalainen lab.

Breast cancer diagnostics is hampered by lack of trained breast pathologists and an ever increasing workload. The consequence is severe regional differences in diagnosis, poor reproducibility of treatment-guiding biomarkers and unacceptable long time to diagnosis. The current AI revolution in combination with digitization of Swedish pathology labs open up a possibility to dramatically improve healthcare.

By training AI-models on world-unique Swedish image datasets we develop AI-models for computational pathology. We focus on several aspects of image analysis such as detection of invasive cancer and premalignant lesions in routine stained sections, histological grading and prognostication. Together, we have access to powerful computational resources necessary to perform the training and validation of the AI methods as well as high-throughput scanning.

Figure. Deep learning (B) versus traditional machine learning (A). In traditional machine learning/image analysis hand-crafted measurements are extracted from each part. Deep learning on the other hand is an end-to-end approach that takes raw images as input and directly learns a model to produce the desired output. Image from Robertson and Hartman, Transl. Res 2017.

Copyright by Hartman Lab. All rights reserved.