Impact of Image Bit Depth Reduction on Deep Learning Performance in Chest Radiograph Analysis: A Multi-institutional Study.
Authors
Affiliations (7)
Affiliations (7)
- Department of Diagnostic and Interventional Radiology, Graduate School of Medicine, Osaka Metropolitan University, Osaka, Japan (H.T., Y.M., T.S., and D.U.); Department of Artificial Intelligence, Graduate School of Medicine, Osaka Metropolitan University, Osaka, Japan (H.T., Y.M., S.L.W. and D.U.).
- Department of Artificial Intelligence, Graduate School of Medicine, Osaka Metropolitan University, Osaka, Japan (H.T., Y.M., S.L.W. and D.U.).
- Center for Digital Transformation of Health Care, Graduate School of Medicine, Kyoto University, Kyoto, Japan (K.S.).
- Department of Diagnostic and Interventional Radiology, Graduate School of Medicine, Osaka Metropolitan University, Osaka, Japan (H.T., Y.M., T.S., and D.U.).
- 3D Imaging Research, Department of Radiology, Massachusetts General Hospital and Harvard Medical School, Boston, MA, United States (M.O.).
- Department of Radiology and Research Institute of Radiology, Asan Medical Center, University of Ulsan College of Medicine, Seoul, Republic of Korea (C.H.S.).
- Department of Diagnostic and Interventional Radiology, Graduate School of Medicine, Osaka Metropolitan University, Osaka, Japan (H.T., Y.M., T.S., and D.U.); Department of Artificial Intelligence, Graduate School of Medicine, Osaka Metropolitan University, Osaka, Japan (H.T., Y.M., S.L.W. and D.U.); Center for Health Science Innovation, Osaka Metropolitan University, Osaka, Japan (D.U.). Electronic address: [email protected].
Abstract
Medical imaging typically generates 12-bit to 16-bit formats, yet conversion to 8-bit is often required. While deep learning has been widely explored in medical imaging, the influence of image bit depth on model performance is not fully understood. This study evaluates the impact of conversion from 16-bit to 8-bit for sex, age, and obesity classification using deep learning. In this retrospective, multi-institutional study, we analyzed 100,002 chest radiographs from 48,047 participants across three institutions. Images were uniformly quantized from the fixed 0-65535 range to 256 levels (0-255) with no additional processing. Three convolutional neural network architectures (ResNet52, EfficientNetB2, and ConvNeXtSmall) were trained on both 16-bit and 8-bit versions of the images. Model performance was evaluated using internal test datasets, randomly split multiple times, and an external test dataset. Statistical analysis included paired comparisons of area under the receiver operating characteristic curve (AUC-ROC) values, with Bonferroni correction for multiple comparisons. Across all architectures and classification tasks, differences between 16-bit and 8-bit model performance were minimal (mean differences ranging from -0.218%-0.184%). Statistical analyses revealed no significant differences in AUC-ROC values between bit depths for any model-task combination (all p values > 0.05 after Bonferroni correction). Effect sizes were small to moderate (Cohen's d ranging from -0.415-0.391). No statistically significant differences were detected between models trained on 16-bit and 8-bit images in the three evaluated tasks. These findings suggest that conversion to 8-bit images may reduce data storage requirements and improve data processing efficiency. Further research is needed to assess their usefulness in clinical tasks.