AnaXNet: Anatomy Aware Multi-label Finding Classification in Chest X-Ray

Agu, Nkechinyere N.; Wu, Joy T.; Chao, Hanqing; Lourentzou, Ismini; Sharma, Arjun; Moradi, Mehdi; Yan, Pingkun; Hendler, James A.

doi:10.1007/978-3-030-87240-3_77

Cited by 22 publications

(9 citation statements)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The image encoders evaluated for global pathology detection perform similarly with an average AUC of 83%. The DenseNet169 performance reported in [1] could not be reached, which we assume might be due to our limited image resolution of 224 × 224. Neither FSRG nor the naïve transfer learning baseline reach the AUC of AnaXNet, which is built upon an object detector and uses features extracted from high-resolution crops.…”

Section: Few-shot Structured Report Generationmentioning

confidence: 86%

“…All other models were trained on a single NVIDIA A40. The DenseNet and Vision Transformer classifiers were trained for 25 epochs with the same hyperparameters as in [1]: Adam optimizer, a learning rate of 1e-4 and unweighted binary cross-entropy loss. The FSRG models were fine-tuned for 10 epochs with a learning rate of 1e-4, AdamW optimizer with no weight decay, and learning rate scheduler with cosine annealing decay and 1 epoch linear warmup.…”

Section: Methodsmentioning

confidence: 99%

“…The first task is localizing pathologies in chest radiographs using the MIMIC-CXR dataset. For this we use the labels provided by ImaGenome [28], that have also been used in [1]. The 9 attributes are: Lung Opacity, Pleural Effusion, Atelectasis Enlarged Cardiac Silhouette, Pulmonary Edema/Hazy Opacity, Pneumothorax, Consolidation, Fluid Overload/Heart Failure and Pneumonia.…”

Section: Localized Pathology Detectionmentioning

confidence: 99%

“…By joining the attribute and location with "in the" we create the template sentences used as an initialization of the classifier, for example "consolidation in the left lung". Following [1] we calculate the area under the receiver operating characteristic curve (AUC) for all possible locations of each pathology and average them to one location-sensitive AUC per pathology. For fair comparison with the image encoder used in CLIP, the DenseNet121 used as a baseline for detection-free localization has been pretrained on the global pathology classification.…”

Section: Localized Pathology Detectionmentioning

confidence: 99%

See 3 more Smart Citations

Few-shot Structured Radiology Report Generation Using Natural Language Prompts

Keicher¹,

Mullakaeva²,

Czempiel³

et al. 2022

Preprint

View full text Add to dashboard Cite

Chest radiograph reporting is time-consuming, and numerous solutions to automate this process have been proposed. Due to the complexity of medical information, the variety of writing styles, and free text being prone to typos and inconsistencies, the efficacy quantifying the clinical accuracy of free-text reports using natural language processing measures is challenging. On the other hand, structured reports ensure consistency and can more easily be used as a quality assurance tool.To accomplish this, we present a strategy for predicting clinical observations and their anatomical location that is easily extensible to other structured findings. First, we train a contrastive language-image model using related chest radiographs and free-text radiological reports. Then, we create textual prompts for each structured finding and optimize a classifier for predicting clinical findings and their associations within the medical image. The results indicate that even when only a few image-level annotations are used for training, the method can localize pathologies in chest radiographs and generate structured reports.

show abstract

Section: Few-shot Structured Report Generationmentioning

confidence: 86%

Section: Methodsmentioning

confidence: 99%

Section: Localized Pathology Detectionmentioning

confidence: 99%

Section: Localized Pathology Detectionmentioning

confidence: 99%

See 2 more Smart Citations

Few-shot Structured Radiology Report Generation Using Natural Language Prompts

Keicher¹,

Mullakaeva²,

Czempiel³

et al. 2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Given the initial training set of anatomical region representations {(R i , R i )} N i=1 , we define a normalized adjacency matrix A ∈ R 2k×2k that captures intra-image and inter-image region correlations. The intra-image correlations corresponding to the two k × k diagonal blocks of A are constructed based on the region-disease co-occurrence [1], i.e., the number of times two anatomical regions co-occur with the same disease or finding in the set of images R i , R i , i = 1, . .…”

Section: Methodsmentioning

confidence: 99%

CheXRelNet: An Anatomy-Aware Model for Tracking Longitudinal Relationships Between Chest X-Rays

Karwande

Mbakwe

et al. 2022

Lecture Notes in Computer Science

Self Cite

View full text Add to dashboard Cite

Despite the progress in utilizing deep learning to automate chest radiograph interpretation and disease diagnosis tasks, change between sequential Chest X-rays (CXRs) has received limited attention. Monitoring the progression of pathologies that are visualized through chest imaging poses several challenges in anatomical motion estimation and image registration, i.e., spatially aligning the two images and modeling temporal dynamics in change detection. In this work, we propose CheXRelNet, a neural model that can track longitudinal pathology change relations between two CXRs. CheXRelNet incorporates local and global visual features, utilizes inter-image and intra-image anatomical information, and learns dependencies between anatomical region attributes, to accurately predict disease change for a pair of CXRs. Experimental results on the Chest ImaGenome dataset show increased downstream performance compared to baselines. Code is available at https://github.com/PLAN-Lab/ChexRelNet.

show abstract