Deep reinforcement learning to detect brain lesions on MRI: a\n proof-of-concept application of reinforcement learning to medical images

2020/08/06 by Joseph N. Stember, Stember, Joseph, Hrithwik Shalu +1 · 1 citation
Computer Science · Medicine · Neuroscience · #AI in cancer detection #Advanced Neural Network Applications #Artificial Intelligence (cs.AI) #Brain Tumor Detection and Classification #FOS: Computer and information sciences #Radiomics and Machine Learning in Medical Imaging

paper · pdf · doi:10.48550/arxiv.2008.02708

openalex publication_date 2020/08/06 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28

Abstract

Purpose: AI in radiology is hindered chiefly by: 1) Requiring large annotated\ndata sets. 2) Non-generalizability that limits deployment to new scanners /\ninstitutions. And 3) Inadequate explainability and interpretability. We believe\nthat reinforcement learning can address all three shortcomings, with robust and\nintuitive algorithms trainable on small datasets. To the best of our knowledge,\nreinforcement learning has not been directly applied to computer vision tasks\nfor radiological images. In this proof-of-principle work, we train a deep\nreinforcement learning network to predict brain tumor location.\n Materials and Methods: Using the BraTS brain tumor imaging database, we\ntrained a deep Q network on 70 post-contrast T1-weighted 2D image slices. We\ndid so in concert with image exploration, with rewards and punishments designed\nto localize lesions. To compare with supervised deep learning, we trained a\nkeypoint detection convolutional neural network on the same 70 images. We\napplied both approaches to a separate 30 image testing set.\n Results: Reinforcement learning predictions consistently improved during\ntraining, whereas those of supervised deep learning quickly diverged.\nReinforcement learning predicted testing set lesion locations with 85%\naccuracy, compared to roughly 7% accuracy for the supervised deep network.\n Conclusion: Reinforcement learning predicted lesions with high accuracy,\nwhich is unprecedented for such a small training set. We believe that\nreinforcement learning can propel radiology AI well past the inherent\nlimitations of supervised deep learning, with more clinician-driven research\nand finally toward true clinical applicability.\n

Citations

• The Multimodal Brain Tumor Image Segmentation Benchmark (BRATS)
• Playing Atari with Deep Reinforcement Learning
• Current Applications and Future Impact of Machine Learning in Radiology
• Gaze2Segment: A Pilot Study for Integrating Eye-Tracking Technology into\n Medical Image Segmentation

Cited by

• Policy to Assist Iteratively Local Segmentation: Optimising Modality and Location Selection for Prostate Cancer Localisation

Related

• Reinforcement learning using Deep Q Networks and Q learning accurately\n localizes brain tumors on MRI with very small training sets
• Deep reinforcement learning-based image classification achieves perfect testing set accuracy for MRI brain tumors with a training set of only 30 images
• Unsupervised deep clustering and reinforcement learning can accurately segment MRI brain tumors with very small training sets
• Deep reinforcement learning with automated label extraction from clinical reports accurately classifies 3D MRI brain volumes
• Deep reinforcement learning in medical imaging: A literature review