TY - GEN
T1 - Barrett's Lesion Detection using a minimal Integer-based Neural Network for Embedded Systems Integration
AU - Boers, Tim G. W.
AU - Kusters, Carolus H. J.
AU - Fockens, Kiki N.
AU - Jukema, Jelmer B.
AU - Jong, Martijn R.
AU - de Groof, Jeroen
AU - Bergman, Jacques J.
AU - van der Sommen, Fons
AU - de With, Peter H. N.
N1 - Funding Information: We gratefully acknowledge the research support provided by Olympus Corporation, Tokyo, Japan. Publisher Copyright: © 2023 SPIE.
PY - 2023
Y1 - 2023
N2 - Embedded processing architectures are often integrated into devices to develop novel functions in a cost-effective medical system. In order to integrate neural networks in medical equipment, these models require specialized optimizations for preparing their integration in a high-efficiency and power-constrained environment. In this paper, we research the feasibility of quantized networks with limited memory for the detection of Barrett's neoplasia. An Efficientnet-lite1+Deeplabv3 architecture is proposed, which is trained using a quantization-aware training scheme, in order to achieve an 8-bit integer-based model. The performance of the quantized model is comparable with float32 precision models. We show that the quantized model with only 5-MB memory is capable of reaching the same performance scores with 95% Area Under the Curve (AUC), compared to a full-precision U-Net architecture, which is 10× larger. We have also optimized the segmentation head for efficiency and reduced the output to a resolution of 32×32 pixels. The results show that this resolution captures sufficient segmentation detail to reach a DICE score of 66.51%, which is comparable to the full floating-point model. The proposed lightweight approach also makes the model quite energy-efficient, since it can be real-time executed on a 2-Watt Coral Edge TPU. The obtained low power consumption of the lightweight Barrett's esophagus neoplasia detection and segmentation system enables the direct integration into standard endoscopic equipment.
AB - Embedded processing architectures are often integrated into devices to develop novel functions in a cost-effective medical system. In order to integrate neural networks in medical equipment, these models require specialized optimizations for preparing their integration in a high-efficiency and power-constrained environment. In this paper, we research the feasibility of quantized networks with limited memory for the detection of Barrett's neoplasia. An Efficientnet-lite1+Deeplabv3 architecture is proposed, which is trained using a quantization-aware training scheme, in order to achieve an 8-bit integer-based model. The performance of the quantized model is comparable with float32 precision models. We show that the quantized model with only 5-MB memory is capable of reaching the same performance scores with 95% Area Under the Curve (AUC), compared to a full-precision U-Net architecture, which is 10× larger. We have also optimized the segmentation head for efficiency and reduced the output to a resolution of 32×32 pixels. The results show that this resolution captures sufficient segmentation detail to reach a DICE score of 66.51%, which is comparable to the full floating-point model. The proposed lightweight approach also makes the model quite energy-efficient, since it can be real-time executed on a 2-Watt Coral Edge TPU. The obtained low power consumption of the lightweight Barrett's esophagus neoplasia detection and segmentation system enables the direct integration into standard endoscopic equipment.
KW - Barrett's neoplasia detection
KW - Embedded systems
KW - full-integer quantization
UR - http://www.scopus.com/inward/record.url?scp=85160213679&partnerID=8YFLogxK
U2 - https://doi.org/10.1117/12.2653890
DO - https://doi.org/10.1117/12.2653890
M3 - Conference contribution
VL - 12465
T3 - Progress in Biomedical Optics and Imaging - Proceedings of SPIE
BT - Medical Imaging 2023
A2 - Iftekharuddin, Khan M.
A2 - Chen, Weijie
PB - SPIE
T2 - Medical Imaging 2023: Computer-Aided Diagnosis
Y2 - 19 February 2023 through 23 February 2023
ER -