Scholarlink

Even today, lung cancer continues to be an important health problem around the world, underlining the need for effective detection methods crucial for increasing the survival rate. Providing lung cancer detection using deep learning from CT scans has been diagnosed in the previous work due to technical constraints, so this work focuses on achieving model efficiency, observability, scalability, and overall CT scan computation scalability. The architecture is built on the Keras framework with automatic differentiation, supports GPU model training, deployment, and inference through KServe in the Kubernetes cluster, can be transformed into compressed SavedModels on TensorFlow, and remotely served via REST/gRPC APIs with auto-scaling and version-controlled model inference. Moreover, observability of system performance, resource utilization, latency percentiles (P50, P90, P99), and system performance with regard to ultra-scalable microservices is executed by the observability layer created with Prometheus, Grafana, Jaeger, and Loki. The infrastructure presented helps combine experimental deep learning and neuroscience with real-life clinics without auditability, fail-with-inform apologies, and perfect explainable operational visibility crucial for AI in healthcare adoption.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Efficient KServe-Based Deep Learning Pipeline for Lung Cancer Detection with Enhanced Observability

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Anupama Babu,
  • Sudheep Elayidom,
  • M. S. Athiramol,
  • Sheenamol Yousaf,
  • Midhun P. Mathew,
  • K. M. Abubeker

摘要

Even today, lung cancer continues to be an important health problem around the world, underlining the need for effective detection methods crucial for increasing the survival rate. Providing lung cancer detection using deep learning from CT scans has been diagnosed in the previous work due to technical constraints, so this work focuses on achieving model efficiency, observability, scalability, and overall CT scan computation scalability. The architecture is built on the Keras framework with automatic differentiation, supports GPU model training, deployment, and inference through KServe in the Kubernetes cluster, can be transformed into compressed SavedModels on TensorFlow, and remotely served via REST/gRPC APIs with auto-scaling and version-controlled model inference. Moreover, observability of system performance, resource utilization, latency percentiles (P50, P90, P99), and system performance with regard to ultra-scalable microservices is executed by the observability layer created with Prometheus, Grafana, Jaeger, and Loki. The infrastructure presented helps combine experimental deep learning and neuroscience with real-life clinics without auditability, fail-with-inform apologies, and perfect explainable operational visibility crucial for AI in healthcare adoption.