eduzhai > Applied Sciences > Engineering >

CorrSigNet Learning CORRelated Prostate Cancer SIGnatures from Radiology and Pathology Images for Improved Computer Aided Diagnosis

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 10 pages

Abstract: Magnetic Resonance Imaging (MRI) is widely used for screening and stagingprostate cancer. However, many prostate cancers have subtle features which arenot easily identifiable on MRI, resulting in missed diagnoses and alarmingvariability in radiologist interpretation. Machine learning models have beendeveloped in an effort to improve cancer identification, but current modelslocalize cancer using MRI-derived features, while failing to consider thedisease pathology characteristics observed on resected tissue. In this paper,we propose CorrSigNet, an automated two-step model that localizes prostatecancer on MRI by capturing the pathology features of cancer. First, the modellearns MRI signatures of cancer that are correlated with correspondinghistopathology features using Common Representation Learning. Second, the modeluses the learned correlated MRI features to train a Convolutional NeuralNetwork to localize prostate cancer. The histopathology images are used only inthe first step to learn the correlated features. Once learned, these correlatedfeatures can be extracted from MRI of new patients (without histopathology orsurgery) to localize cancer. We trained and validated our framework on a uniquedataset of 75 patients with 806 slices who underwent MRI followed byprostatectomy surgery. We tested our method on an independent test set of 20prostatectomy patients (139 slices, 24 cancerous lesions, 1.12M pixels) andachieved a per-pixel sensitivity of 0.81, specificity of 0.71, AUC of 0.86 anda per-lesion AUC of $0.96 pm 0.07$, outperforming the current state-of-the-artaccuracy in predicting prostate cancer using MRI.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...