CheSS: Chest X-Ray Pre-trained Model via Self-supervised Contrastive Learning

Abstract

Training deep learning models on medical images heavily depends on experts’ expensive and laborious manual labels. In addition, these images, labels, and even models themselves are not widely publicly accessible and suffer from various kinds of bias and imbalances. In this paper, chest X-ray pre-trained model via self-supervised contrastive learning (CheSS) was proposed to learn models with various representations in chest radiographs (CXRs). Our contribution is a publicly accessible pretrained model trained with a 4.8-M CXR dataset using self-supervised learning with a contrastive learning and its validation with various kinds of downstream tasks including classification on the 6-class diseases in internal dataset, diseases classification in CheXpert, bone suppression, and nodule generation. When compared to a scratch model, on the 6-class classification test dataset, we achieved 28.5% increase in accuracy. On the CheXpert dataset, we achieved 1.3% increase in mean area under the receiver operating characteristic curve on the full dataset and 11.4% increase only using 1% data in stress test manner. On bone suppression with perceptual loss, we achieved improvement in peak signal to noise ratio from 34.99 to 37.77, structural similarity index measure from 0.976 to 0.977, and root-square-mean error from 4.410 to 3.301 when compared to ImageNet pretrained model. Finally, on nodule generation, we achieved improvement in Fréchet inception distance from 24.06 to 17.07. Our study showed the decent transferability of CheSS weights. CheSS weights can help researchers overcome data imbalance, data shortage, and inaccessibility of medical image datasets. CheSS weight is available at https://github.com/mi2rl/CheSS.

Updated Primer on Generative Artificial Intelligence and Large Language Models in Medical Imaging for Medical Professionals

“…One group of researchers trained a vision foundation model on 100 million medical images, including radiographs, CT images, MR images, and ultrasound images [ 67 ]. Another group of researchers trained a self-supervised network on 4.8 million chest radiographs [ 68 ]. However, the networks were not scalable despite these studies training their networks on a vast amount of data and demonstrating their diverse utility.…”

Automatic feature extraction by supervised and contrastive self-supervised learning based on wavelet and hard negatives to detect HIFU lesion area

“…[4][5][6][7][8][9][10][11][12][13] However, medical image processing faces unique challenges, notably limited data and labeling; consequently, recent efforts have aimed to reduce reliance on data annotation across various medical image types [14][15][16][17]. Studies have predominantly focused on self-supervised methods for X-ray images [18][19][20][21][22][23] with comparatively fewer articles addressing MRI [24][25][26][27] and CT-scan [28][29][30][31][32], and notably fewer on ultrasound images. [33,34] Articles discussing HIFU control and monitoring include [35] presenting a method utilizing ultrasound signals as input to a feedforward neural network for lesion area detection.…”

Self-supervised Learning for Chest CT - Training Strategies and Effect on Downstream Applications

“…Simul-taneously, complex vision-language models like CLIP [16], DALL-E [25], FLAVA [26], and Socratec models [34] were trained in a self-supervised way for understanding correlation in multi-modal data, and enabling reasoning across multiple modalities, and even applied in generation tasks. SSL is also being popular for radiology image analysis [3], [6], [20], [27]; however different from natural images, self-supervision using radiology images needs targeted pretraining using unlabeled medical images thus understanding the effects of various self-supervision strategies are important to plan experiments in advance and reduce unnecessary waste of computation resources and time of training. Although there exists a "gap" in literature since no study exists that benchmark SSL strategies for medical images.…”