Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
MATCH: Multi-faceted Adaptive Topo-Consistency for Semi-Supervised Histopathology Segmentation
Authors: Meilong Xu, Xiaoling Hu, Shahira Abousamra, Chen Li, Chao Chen
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on three widely used histopathology image datasets, our method significantly improves the topological accuracy while achieving comparable pixel-wise performance with limited annotations. We conduct comprehensive evaluations on three publicly available histopathology image datasets on both pixel-wise and topology-wise metrics. We benchmark our method against classic and recent state-of-the-art semi-supervised medical image segmentation methods, including MT [57], EM [59], UA-MT [79], URPC [39], XNet [83], PMT [10], and Topo Semi Seg [69]. |
| Researcher Affiliation | Academia | 1Stony Brook University, NY, USA 2 Massachusetts General Hospital and Harvard Medical School, MA, USA 3Department of Biomedical Data Science, Stanford University, CA, USA |
| Pseudocode | No | The paper describes the methodology (e.g., Section 3 and subsections) and illustrates concepts with diagrams (e.g., Figure 4), but it does not include any explicitly labeled 'Pseudocode' or 'Algorithm' blocks, nor does it present structured, code-like procedural steps. |
| Open Source Code | Yes | Code is available at https://github.com/Melon Xu/MATCH. |
| Open Datasets | Yes | We evaluate our proposed method on Colorectal Adenocarcinoma Gland (CRAG) [11], Gland Segmentation in Colon Histology Images Challenge (Gla S) [53], and Multi-Organ Nuclei Segmentation (Mo Nu Seg) [32]. |
| Dataset Splits | Yes | Colorectal Adenocarcinoma Gland (CRAG) [11] consists of 213 hematoxylin and eosin (H&E)-stained colorectal adenocarcinoma image tiles... Officially, the dataset is partitioned into 173 training samples and 40 testing samples. For our experiments, the training subset is further divided into 153 images for model training and 20 images for validation. For semi-supervised scenarios with 10% and 20% labeled data, we randomly select 16 and 31 labeled images, respectively, for training. Gland Segmentation in Colon Histology Images Challenge (Gla S) [53]... The official split includes 85 training images and 80 testing images. In our experimental setup, the training set is divided into 68 images for model training and 17 for validation. We randomly select 7 and 14 labeled images to represent 10% and 20% of labeled training data scenarios, respectively. Multi-Organ Nuclei Segmentation (Mo Nu Seg) [32] dataset... Officially, it consists of 30 training images... and 14 images designated for testing. For our experiments, we reserve 20% (6 images) of the training set for validation. In experiments involving 10% and 20% labeled data splits, we randomly select 3 and 5 labeled images, respectively, for training. |
| Hardware Specification | Yes | The experiments are conducted on an NVIDIA RTX A6000 GPU (48 GB), using a 24-core Intel Xeon Gold 6248R CPU @ 3.00 GHz and 192 GB RAM. |
| Software Dependencies | No | All training is implemented using Py Torch [47] and optimized using the Adam optimizer [28]. The paper mentions PyTorch but does not provide a specific version number for it or any other key software libraries. |
| Experiment Setup | Yes | Training hyperparameters are set as follows: the batch size is 16 and the learning rate is 5 10 4. Both labeled and unlabeled data undergo pre-processing through random cropping (with cropping size of 256 256), followed by data augmentation procedures including random rotation and flipping as weak augmentations, and color adjustments and morphological shifts for stronger augmentations. ... The EMA decay rate α is set to 0.999. Within the supervised loss, the weights assigned to the cross-entropy loss and Dice loss are equally set to 0.5. The weight of the pixel-wise consistency loss is calculated by the Gaussian ramp-up function λcons = k e 5 (1 τ T )2, where k = 0.1 and T is the total number of iterations. Additionally, λintra and λtemp are both set to 0.001. |