Leveraging Other Datasets for Medical Imaging Classification: Evaluation of Transfer, Multi-task and Semi-supervised Learning

Shang, Hong; Sun, Zhongqian; Yang, Wei; Fu, Xinghui; Zheng, Han; Jia, Chang; Huang, Junzhou

doi:10.1007/978-3-030-32254-0_48

Cited by 24 publications

(17 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our method is built on the MTL strategy (45) previously studies in the context of other medical imaging applications (46,47). State-of-the-art methods often implement MTL as a combination of detection and classification tasks (37,(48)(49)(50).…”

Section: Related Workmentioning

confidence: 99%

Leveraging Multi-Task Learning to Cope With Poor and Missing Labels of Mammograms

Tardy¹,

Mateus²

2022

Front. Radio

View full text Add to dashboard Cite

In breast cancer screening, binary classification of mammograms is a common task aiming to determine whether a case is malignant or benign. A Computer-Aided Diagnosis (CADx) system based on a trainable classifier requires clean data and labels coming from a confirmed diagnosis. Unfortunately, such labels are not easy to obtain in clinical practice, since the histopathological reports of biopsy may not be available alongside mammograms, while normal cases may not have an explicit follow-up confirmation. Such ambiguities result either in reducing the number of samples eligible for training or in a label uncertainty that may decrease the performances. In this work, we maximize the number of samples for training relying on multi-task learning. We design a deep-neural-network-based classifier yielding multiple outputs in one forward pass. The predicted classes include binary malignancy, cancer probability estimation, breast density, and image laterality. Since few samples have all classes available and confirmed, we propose to introduce the uncertainty related to the classes as a per-sample weight during training. Such weighting prevents updating the network's parameters when training on uncertain or missing labels. We evaluate our approach on the public INBreast and private datasets, showing statistically significant improvements compared to baseline and independent state-of-the-art approaches. Moreover, we use mammograms from Susan G. Komen Tissue Bank for fine-tuning, further demonstrating the ability to improve the performances in our multi-task learning setup from raw clinical data. We achieved the binary classification performance of AUC = 80.46 on our private dataset and AUC = 85.23 on the INBreast dataset.

show abstract

Section: Related Workmentioning

confidence: 99%

Leveraging Multi-Task Learning to Cope With Poor and Missing Labels of Mammograms

Tardy¹,

Mateus²

2022

Front. Radio

View full text Add to dashboard Cite

show abstract

“…Due to the difficulty of data annotation, SSL is widely used in medical imaging processing, e.g., medical image detection [48,60], classification [18,34,39] and segmentation [36,61]. [48] proposes an SSL method for 3D medical image detection that uses a generic generalization of the focal loss to regularize the SSL training.…”

Section: Related Workmentioning

confidence: 99%

BoostMIS: Boosting Medical Image Semi-supervised Learning with Adaptive Pseudo Labeling and Informative Active Annotation

Zhang¹,

Li²,

Hallinan³

et al. 2022

Preprint

View full text Add to dashboard Cite

In this paper, we propose a novel semi-supervised learning (SSL) framework named BoostMIS that combines adaptive pseudo labeling and informative active annotation to unleash the potential of medical image SSL models: (1) BoostMIS can adaptively leverage the cluster assumption and consistency regularization of the unlabeled data according to the current learning status. This strategy can adaptively generate one-hot "hard" labels converted from task model predictions for better task model training. (2) For the unselected unlabeled images with low confidence, we introduce an Active learning (AL) algorithm to find the informative samples as the annotation candidates by exploiting virtual adversarial perturbation and model's density-aware entropy. These informative candidates are subsequently fed into the next training cycle for better SSL label propagation. Notably, the adaptive pseudo-labeling and informative active annotation form a learning closedloop that are mutually collaborative to boost medical image SSL. To verify the effectiveness of the proposed method, we collected a metastatic epidural spinal cord compression (MESCC) dataset that aims to optimize MESCC diagnosis and classification for improved specialist referral and treatment. We conducted an extensive experimental study of BoostMIS on MESCC dataset. The experimental results verify our framework's effectiveness with a significant improvement over various state-of-the-art methods.

show abstract

“…Again, the labeling/annotating tasks to prepare the raw images of the training data are often performed manually by specialists via visual evaluation [ 32 ], which depends on the person’s skill, and is very laborious, expensive, and time-consuming. Thus, scarce and weakly labeled (annotated) FF-OCT images create a major challenge in training deep convolutional neural networks [ 33 ] to develop computer-aided diagnosis. To address this problem, we attempted to develop a simulation algorithm for the automatic generation of ground truth and the corresponding augmented OCT input images simultaneously.…”

Section: Introductionmentioning

confidence: 99%

Generation of Augmented Capillary Network Optical Coherence Tomography Image Data of Human Skin for Deep Learning and Capillary Segmentation

et al. 2021

View full text Add to dashboard Cite

The segmentation of capillaries in human skin in full-field optical coherence tomography (FF-OCT) images plays a vital role in clinical applications. Recent advances in deep learning techniques have demonstrated a state-of-the-art level of accuracy for the task of automatic medical image segmentation. However, a gigantic amount of annotated data is required for the successful training of deep learning models, which demands a great deal of effort and is costly. To overcome this fundamental problem, an automatic simulation algorithm to generate OCT-like skin image data with augmented capillary networks (ACNs) in a three-dimensional volume (which we called the ACN data) is presented. This algorithm simultaneously acquires augmented FF-OCT and corresponding ground truth images of capillary structures, in which potential functions are introduced to conduct the capillary pathways, and the two-dimensional Gaussian function is utilized to mimic the brightness reflected by capillary blood flow seen in real OCT data. To assess the quality of the ACN data, a U-Net deep learning model was trained by the ACN data and then tested on real in vivo FF-OCT human skin images for capillary segmentation. With properly designed data binarization for predicted image frames, the testing result of real FF-OCT data with respect to the ground truth achieved high scores in performance metrics. This demonstrates that the proposed algorithm is capable of generating ACN data that can imitate real FF-OCT skin images of capillary networks for use in research and deep learning, and that the model for capillary segmentation could be of wide benefit in clinical and biomedical applications.

show abstract

Leveraging Other Datasets for Medical Imaging Classification: Evaluation of Transfer, Multi-task and Semi-supervised Learning

Cited by 24 publications

References 12 publications

Leveraging Multi-Task Learning to Cope With Poor and Missing Labels of Mammograms

Leveraging Multi-Task Learning to Cope With Poor and Missing Labels of Mammograms

BoostMIS: Boosting Medical Image Semi-supervised Learning with Adaptive Pseudo Labeling and Informative Active Annotation

Generation of Augmented Capillary Network Optical Coherence Tomography Image Data of Human Skin for Deep Learning and Capillary Segmentation

Contact Info

Product

Resources

About