arXiv (Cornell University) · 2020 · 11 citations · 0 references
ComputingComputer SciencePathologyBiomedical ImagingComputer VisionMachine LearningBreast PathologyCancer DetectionCancer Early DetectionObject DetectionPattern RecognitionRepresentation LearningBreast Cancer ProgressionBreast Tissue ImagingConvolutional Neural Network DetectionHistopathological InterpretationPixel-level AnnotationSparse Feature Representation
Empirical evaluation of breast tissue biopsies for mitotic nuclei detection\nis considered an important prognostic biomarker in tumor grading and cancer\nprogression. However, automated mitotic nuclei detection poses several\nchallenges because of the unavailability of pixel-level annotations, different\nmorphological configurations of mitotic nuclei, their sparse representation,\nand close resemblance with non-mitotic nuclei. These challenges undermine the\nprecision of the automated detection model and thus make detection difficult in\na single phase. This work proposes an end-to-end detection system for mitotic\nnuclei identification in breast cancer histopathological images. Deep object\ndetection-based Mask R-CNN is adapted for mitotic nuclei detection that\ninitially selects the candidate mitotic region with maximum recall. However, in\nthe second phase, these candidate regions are refined by multi-object loss\nfunction to improve the precision. The performance of the proposed detection\nmodel shows improved discrimination ability (F-score of 0.86) for mitotic\nnuclei with significant precision (0.86) as compared to the two-stage detection\nmodels (F-score of 0.701) on TUPAC16 dataset. Promising results suggest that\nthe deep object detection-based model has the potential to learn the\ncharacteristic features of mitotic nuclei from weakly annotated data and\nsuggests that it can be adapted for the identification of other nuclear bodies\nin histopathological images.\n