Learning to segment images with classification labels

Ciga, Ozan; Martel, Anne L.

doi:10.1016/j.media.2020.101912

Cited by 18 publications

(17 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…21 We did not find any other publication mentioning these mistakes explicitly. The ground truth used by others varied between the STAPLE consensus, 22 a simple majority vote, 23 using a single expert 24 or merging the different annotations into a "grade probability map". 25…”

Section: Resultsmentioning

confidence: 99%

“…Expert 1 appears to be an outlier, which is interesting to note because this expert was used as a single source of ground truth in one of the publications using the dataset. 24 Such visualization provides an interesting opportunity for discussing the results of an algorithm. In challenges such as Gleason2019, the algorithms are ranked based on their similarity to the consensus.…”

Section: Visualizing Expert and Consensus Methods Agreementmentioning

confidence: 99%

See 1 more Smart Citation

Processing multi-expert annotations in digital pathology: a study of the Gleason 2019 challenge

Foucart

Debeir

Decaestecker

2021

17th International Symposium on Medical Information Processing and Analysis

View full text Add to dashboard Cite

Deep learning algorithms rely on large amounts of annotations for learning and testing. In digital pathology, a ground truth is rarely available, and many tasks show large inter-expert disagreement. Using the Gleason2019 dataset, we analyse how the choices we make in getting the ground truth from multiple experts may affect the results and the conclusions we could make from challenges and benchmarks. We show that using undocumented consensus methods, as is often done, reduces our ability to properly analyse challenge results. We also show that taking into account each expert's annotations enriches discussions on results and is more in line with the clinical reality and complexity of the application.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Visualizing Expert and Consensus Methods Agreementmentioning

confidence: 99%

Processing multi-expert annotations in digital pathology: a study of the Gleason 2019 challenge

Foucart

Debeir

Decaestecker

2021

17th International Symposium on Medical Information Processing and Analysis

View full text Add to dashboard Cite

show abstract

“…This is because most of the modern CV systems (that are supervised) try to learn some form of image representations by finding a pattern that links the data points to their respective annotations in large datasets [3]. Moreover, the data annotation efforts vary from task to task, and it is estimated that the time spent on image segmentation and object detection (i.e., carefully drawing boundaries) is four times longer than the image classification itself [4]. The annotation efforts become significantly higher when it comes to highly regulated and specialized domains like medicine and finance in which the expertise level of the human annotator matters more than in any other domain.…”

Section: Introductionmentioning

confidence: 99%

Saliency Can Be All You Need In Contrastive Self-Supervised Learning

Kocaman¹,

Shir²,

Bäck³

et al. 2022

Preprint

View full text Add to dashboard Cite

We propose an augmentation policy for Contrastive Self-Supervised Learning (SSL) in the form of an already established Salient Image Segmentation technique entitled Global Contrast based Salient Region Detection. This detection technique, which had been devised for unrelated Computer Vision tasks, was empirically observed to play the role of an augmentation facilitator within the SSL protocol. This observation is rooted in our practical attempts to learn, by SSL-fashion, aerial imagery of solar panels, which exhibit challenging boundary patterns. Upon the successful integration of this technique on our problem domain, we formulated a generalized procedure and conducted a comprehensive, systematic performance assessment with various Contrastive SSL algorithms subject to standard augmentation techniques. This evaluation, which was conducted across multiple datasets, indicated that the proposed technique indeed contributes to SSL. We hypothesize whether salient image segmentation may suffice as the only augmentation policy in Contrastive SSL when treating downstream segmentation tasks.

show abstract

“…tumor segmentation), this price could be higher considering the hourly wage of a radiologist. As for the time, (Ciga and Martel, 2021) reports that it takes between 15 minutes and two hours depending on the size and resolution to segment a single image of lymph nodes for breast cancer. An approach dealing with missing modalities and requiring less labels can reduce the monetary and time-related costs.…”

Section: Introductionmentioning

confidence: 99%

Benefits of Linear Conditioning with Metadata for Image Segmentation

Lemay,

Gros,

Vincent

et al. 2021

Preprint

View full text Add to dashboard Cite

Medical images are often accompanied by metadata describing the image (vendor, acquisition parameters) and the patient (disease type or severity, demographics, genomics). This metadata is usually disregarded by image segmentation methods. In this work, we adapt a linear conditioning method called FiLM (Feature-wise Linear Modulation) for image segmentation tasks. This FiLM adaptation enables integrating metadata into segmentation models for better performance. We observed an average Dice score increase of 5.1% on spinal cord tumor segmentation when incorporating the tumor type with FiLM. The metadata modulates the segmentation process through low-cost affine transformations applied on feature maps which can be included in any neural network's architecture. Additionally, we assess the relevance of segmentation FiLM layers for tackling common challenges in medical imaging: training with limited or unbalanced number of annotated data, multi-class training with missing segmentations, and model adaptation to multiple tasks. Our results demonstrated the following benefits of FiLM for segmentation: FiLMed U-Net was robust to missing labels and reached higher Dice scores with few labels (up to 16.7%) compared to single-task U-Net. The code is open-source and available at www.ivadomed.org.

show abstract

Learning to segment images with classification labels

Cited by 18 publications

References 10 publications

Processing multi-expert annotations in digital pathology: a study of the Gleason 2019 challenge

Processing multi-expert annotations in digital pathology: a study of the Gleason 2019 challenge

Saliency Can Be All You Need In Contrastive Self-Supervised Learning

Benefits of Linear Conditioning with Metadata for Image Segmentation

Contact Info

Product

Resources

About