OMG-Net: A Deep Learning Framework Deploying Segment Anything to Detect Pan-Cancer Mitotic Figures from Haematoxylin and Eosin-Stained Slides
Authors:
Zhuoyan Shen,
Mikael Simard,
Douglas Brand,
Vanghelita Andrei,
Ali Al-Khader,
Fatine Oumlil,
Katherine Trevers,
Thomas Butters,
Simon Haefliger,
Eleanna Kara,
Fernanda Amary,
Roberto Tirabosco,
Paul Cool,
Gary Royle,
Maria A. Hawkins,
Adrienne M. Flanagan,
Charles-Antoine Collins Fekete
Abstract:
Mitotic activity is an important feature for grading several cancer types. Counting mitotic figures (MFs) is a time-consuming, laborious task prone to inter-observer variation. Inaccurate recognition of MFs can lead to incorrect grading and hence potential suboptimal treatment. In this study, we propose an artificial intelligence (AI)-aided approach to detect MFs in digitised haematoxylin and eosi…
▽ More
Mitotic activity is an important feature for grading several cancer types. Counting mitotic figures (MFs) is a time-consuming, laborious task prone to inter-observer variation. Inaccurate recognition of MFs can lead to incorrect grading and hence potential suboptimal treatment. In this study, we propose an artificial intelligence (AI)-aided approach to detect MFs in digitised haematoxylin and eosin-stained whole slide images (WSIs). Advances in this area are hampered by the limited number and types of cancer datasets of MFs. Here we establish the largest pan-cancer dataset of mitotic figures by combining an in-house dataset of soft tissue tumours (STMF) with five open-source mitotic datasets comprising multiple human cancers and canine specimens (ICPR, TUPAC, CCMCT, CMC and MIDOG++). This new dataset identifies 74,620 MFs and 105,538 mitotic-like figures. We then employed a two-stage framework (the Optimised Mitoses Generator Network (OMG-Net) to classify MFs. The framework first deploys the Segment Anything Model (SAM) to automate the contouring of MFs and surrounding objects. An adapted ResNet18 is subsequently trained to classify MFs. OMG-Net reaches an F1-score of 0.84 on pan-cancer MF detection (breast carcinoma, neuroendocrine tumour and melanoma), largely outperforming the previous state-of-the-art MIDOG++ benchmark model on its hold-out testing set (e.g. +16% F1-score on breast cancer detection, p<0.001) thereby providing superior accuracy in detecting MFs on various types of tumours obtained with different scanners.
△ Less
Submitted 17 July, 2024;
originally announced July 2024.
Mixed Linear Layouts: Complexity, Heuristics, and Experiments
Authors:
Philipp de Col,
Fabian Klute,
Martin Nöllenburg
Abstract:
A $k$-page linear graph layout of a graph $G = (V,E)$ draws all vertices along a line $\ell$ and each edge in one of $k$ disjoint halfplanes called pages, which are bounded by $\ell$. We consider two types of pages. In a stack page no two edges should cross and in a queue page no edge should be nested by another edge. A crossing (nesting) in a stack (queue) page is called a conflict. The algorithm…
▽ More
A $k$-page linear graph layout of a graph $G = (V,E)$ draws all vertices along a line $\ell$ and each edge in one of $k$ disjoint halfplanes called pages, which are bounded by $\ell$. We consider two types of pages. In a stack page no two edges should cross and in a queue page no edge should be nested by another edge. A crossing (nesting) in a stack (queue) page is called a conflict. The algorithmic problem is twofold and requires to compute (i) a vertex ordering and (ii) a page assignment of the edges such that the resulting layout is either conflict-free or conflict-minimal. While linear layouts with only stack or only queue pages are well-studied, mixed $s$-stack $q$-queue layouts for $s,q \ge 1$ have received less attention. We show NP-completeness results on the recognition problem of certain mixed linear layouts and present a new heuristic for minimizing conflicts. In a computational experiment for the case $s, q = 1$ we show that the new heuristic is an improvement over previous heuristics for linear layouts.
△ Less
Submitted 23 August, 2019;
originally announced August 2019.
