Biased Data, Biased AI: Deep Networks Predict the Acquisition Site of TCGA Images

Dehkharghanian, Taher; Bidgoli, Azam Asilian; Riasatian, Abtin; Mazaheri, Pooria; Campbell, Clinton J. V.; Pantanowitz, Liron; Tizhoosh, H.R.; Rahnamayan, Shahryar

doi:10.21203/rs.3.rs-943804/v1

Cited by 10 publications

(17 citation statements)

References 19 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The experiments showed that the model gave a remarkable performance, and provide important assistance for health experts and diagnostic radiographers. The whole-slide images available in the Cancer Genome Atlas (TCGA) have been used by Dehkharghanian et al [15] for developing a deep-learning model for the automatic analysis of tumor slides where in order to recognize tumor in lung versus normal tissue. After testing and evaluating the deep-learning model by the training they found that artificial intelligence technology based on deep-learning models can help pathologists in cancer subtype detection or gene mutations in any cancer type with a save of costs and time.…”

Section: Introductionmentioning

confidence: 99%

Lung cancer detection using image processing and deep learning

Al-Ameer

Hussien

Al-Ameri

2022

IJEECS

View full text Add to dashboard Cite

This project is about the detection of lung cancer by training a model of deep neural networks using histopathological lung cancer tissue images. Deferent models have been proposed for detecting lung cancer cells automatically involving Inception V3, Random Forest, and convolutional neural network (CNN). The deep convolutional neural network has been trained to extract important features that facilitate build detection and diagnosis of lung cancer cells more efficiently and accurately. The proposed method in this project has accomplished promising and satisfactory results in terms of accuracy, precision, recall, F-score, and specificity measure in lung cancer detection. Furthermore, it has been applied on dataset which contains 178,000 photos. The accuracy values that are obtained are accuracy 97.09%, precision 96.89%, recall 97.31%, F-score measure 97.09%, and specificity measure 96.88%.

show abstract

Section: Introductionmentioning

confidence: 99%

Lung cancer detection using image processing and deep learning

Al-Ameer

Hussien

Al-Ameri

2022

IJEECS

View full text Add to dashboard Cite

show abstract

“…The discrepancy between the source and target domain distributions is known as domain shift [7]. According to Yagi et al [8], the domain shift in histopathology usually occurs when WSIs are acquired at different trial sites due to differences in slide preparation, staining procedure, and scanner characteristics as well as biases in developing AI models [9]. Overall, correct diagnosis is a critical task in histopathology which can directly influence the treatment outcome.…”

Section: Introductionmentioning

confidence: 99%

Hospital-Agnostic Image Representation Learning in Digital Pathology

Sikaroudi

Rahnamayan

Tizhoosh

2022

2022 44th Annual International Conference of the IEEE Engineering in Medicine &Amp; Biology Society (EMBC)

Self Cite

View full text Add to dashboard Cite

Whole Slide Images (WSIs) in digital pathology are used to diagnose cancer subtypes. The difference in procedures to acquire WSIs at various trial sites gives rise to variability in the histopathology images, thus making consistent diagnosis challenging. These differences may stem from variability in image acquisition through multi-vendor scanners, variable acquisition parameters, and differences in staining procedure; as well, patient demographics may bias the glass slide batches before image acquisition. These variabilities are assumed to cause a domain shift in the images of different hospitals. It is crucial to overcome this domain shift because an ideal machine-learning model must be able to work on the diverse sources of images, independent of the acquisition center. A domain generalization technique is leveraged in this study to improve the generalization capability of a Deep Neural Network (DNN), to an unseen histopathology image set (i.e., from an unseen hospital/trial site) in the presence of domain shift. According to experimental results, the conventional supervisedlearning regime generalizes poorly to data collected from different hospitals. However, the proposed hospital-agnostic learning can improve the generalization considering the lowdimensional latent space representation visualization, and classification accuracy results.

show abstract

“…That is possibly related to the non-optimal design of the DNNs structure which leading to extracted non-optimal set of features [23]. Consequently, it seems that DNNs have a tendency to learn irrelevant shortcuts patterns related to data acquisition sites, rather than the actual underlying morphological information [19].…”

Section: Introductionmentioning

confidence: 99%

“…In a recent study [19], the existence of bias in histopathology images of The Cancer Genome Atlas (TCGA) [20] was investigated. It was established that the deep features extracted from the images are able to accurately distinguish the WSIs based on their acquisition site.…”

Section: Introductionmentioning

confidence: 99%

Bias Reduction in Representation of Histopathology Images Using Deep Feature Selection

Bidgoli

Rahnamayan

Dehkharghanian

et al. 2022

Preprint

Self Cite

View full text Add to dashboard Cite

Appearing traces of bias in deep networks is a serious issue that can play a significant role in ethics and generalization. Recent studies report that the deep features extracted from the histopathology images of The Cancer Genome Atlas (TCGA), the largest publicly available archive of 11,000 patients covering 25 organs and 32 cancer subtypes, are surprisingly able to accurately classify the whole slide images (WSIs) based on their acquisition site. This is clear evidence that the utilized Deep Neural Networks (DNNs) unexpectedly detect the specific patterns of the source site rather than histomorphologic patterns, biased behaviour resulting in degraded generalization. This observation motivated us to propose a method to alleviate the destructive impact of hospital bias through a novel feature selection process. To this effect, we have proposed an evolutionary strategy select a small set of optimal features to not only accurately represent the histological patterns of tissue samples but also to eliminate the features leading to internal bias toward the institution. The performance of the proposed method has been assessed using the features extracted from TCGA images made available by NIH (National Institute of Health). The selected features extracted by a state-of-the-art network trained on TCGA images (i.e., the KimiaNet), considerably decreased the institutional bias. The proposed scheme is not limited to DNNs and/or specific types of biases; it can be employed to reduce various kinds of bias in a much more comprehensive range of data-driven feature extraction models. The conducted experiments, the external validation in specific, clearly demonstrate that the bias cannot be fully controlled just during training a model. Therefore, a feature selection for downstream tasks plays a crucial role in significantly reducing the bias.

show abstract

Biased Data, Biased AI: Deep Networks Predict the Acquisition Site of TCGA Images

Cited by 10 publications

References 19 publications

Lung cancer detection using image processing and deep learning

Lung cancer detection using image processing and deep learning

Hospital-Agnostic Image Representation Learning in Digital Pathology

Bias Reduction in Representation of Histopathology Images Using Deep Feature Selection

Contact Info

Product

Resources

About