Example: Neural network WSI classification

This example performs patch-wise classification of a whole slide microscopy image using a neural network.

/**
 * Examples/Segmentation/neuralNetworkWSISegmentation.cpp
 *
 * If you edit this example, please also update the wiki and source code file in the repository.
 */
#include <FAST/Tools/CommandLineParser.hpp>
#include <FAST/Algorithms/NeuralNetwork/InferenceEngineManager.hpp>
#include <FAST/Visualization/SimpleWindow.hpp>
#include <FAST/Visualization/HeatmapRenderer/HeatmapRenderer.hpp>
#include <FAST/Importers/WholeSlideImageImporter.hpp>
#include <FAST/Visualization/ImagePyramidRenderer/ImagePyramidRenderer.hpp>
#include <FAST/Algorithms/ImagePatch/PatchGenerator.hpp>
#include <FAST/Algorithms/ImagePatch/PatchStitcher.hpp>
#include <FAST/Algorithms/NeuralNetwork/SegmentationNetwork.hpp>
#include <FAST/Algorithms/TissueSegmentation/TissueSegmentation.hpp>

using namespace fast;

int main(int argc, char** argv) {
    Reporter::setGlobalReportMethod(Reporter::COUT);
    CommandLineParser parser("Neural network WSI classification example");
    parser.addChoice("inference-engine",
            {"default", "OpenVINO", "TensorFlowCPU", "TensorFlowCUDA", "TensorRT", "TensorFlowROCm"},
            "default",
            "Which neural network inference engine to use");
    parser.addPositionVariable(1,
            "filename",
            Config::getTestDataPath() + "/WSI/A05.svs",
            "WSI to process");
    parser.parse(argc, argv);

    auto importer = WholeSlideImageImporter::New();
    importer->setFilename(parser.get("filename"));

    auto tissueSegmentation = TissueSegmentation::New();
    tissueSegmentation->setInputConnection(importer->getOutputPort());

    auto generator = PatchGenerator::New();
    generator->setPatchSize(512, 512);
    generator->setPatchLevel(0);
    generator->setInputConnection(importer->getOutputPort());
    generator->setInputConnection(1, tissueSegmentation->getOutputPort());

    auto network = NeuralNetwork::New();
    if(parser.get("inference-engine") != "default") {
        network->setInferenceEngine(parser.get("inference-engine"));
    }
    const auto engine = network->getInferenceEngine()->getName();
    network->setInferenceEngine(engine);
    if(engine.substr(0, 10) == "TensorFlow") {
        network->setOutputNode(0, "sequential/dense_1/Softmax", NodeType::TENSOR);
    } else if(engine == "TensorRT") {
        network->setInputNode(0, "input_1", NodeType::IMAGE, TensorShape{-1, 3, 512, 512});
        network->setOutputNode(0, "sequential/dense_1/Softmax", NodeType::TENSOR, TensorShape{-1, 3});
    }
    network->load(Config::getTestDataPath() + "NeuralNetworkModels/wsi_classification." +
                  network->getInferenceEngine()->getDefaultFileExtension());
    network->setInputConnection(generator->getOutputPort());
    network->setScaleFactor(1.0f / 255.0f);

    auto stitcher = PatchStitcher::New();
    stitcher->setInputConnection(network->getOutputPort());

    auto renderer = ImagePyramidRenderer::New();
    renderer->addInputConnection(importer->getOutputPort());

    auto heatmapRenderer = HeatmapRenderer::New();
    heatmapRenderer->addInputConnection(stitcher->getOutputPort());
    //heatmapRenderer->setMinConfidence(0.5);
    heatmapRenderer->setMaxOpacity(0.3);

    auto window = SimpleWindow::New();
    window->addRenderer(renderer);
    window->addRenderer(heatmapRenderer);
    window->enableMaximized();
    window->set2DMode();
    window->start();
}