🍩 Database of Original & Non-Theoretical Uses of Topology

(found 3 matches in 0.001164s)

  3. Histopathological Cancer Detection With Topological Signatures (2023)

    Ankur Yadav, Faisal Ahmed, Ovidiu Daescu, Reyhan Gedik, Baris Coskunuzer
    Abstract We present a transformative approach to histopathological cancer detection and grading by introducing a very powerful feature extraction method based on the latest topological data analysis tools. By analyzing the evolution of topological patterns in different color channels, we discovered that every tumor class leaves its own topological footprint in histopathological images, allowing to extract feature vectors that can be used to reliably identify tumor classes.Our topological signatures, even when combined with traditional machine learning methods, provide very fast and highly accurate results in various settings. While most DL models work well for one type of cancer, our model easily adapts to different scenarios, and consistently gives highly competitive results with the state-of-the-art models on benchmark datasets across multiple cancer types including bone, colon, breast, cervical (cytopathology), and prostate cancer. Unlike most DL models, our proposed Topo-ML model does not need any data augmentation or pre-processing steps and works perfectly on small datasets. The model is computationally very efficient, with end-to-end processing taking only a few hours for datasets consisting of thousands of images.