Digital pathology begins when a thin section of human tissue is mounted on a glass slide and
scanned into a Whole-Slide Image (WSI) – typically 80,000 x 60,000 pixels and a size of 500MB to
5GB even in compressed form. WSIs drive state-of-the-art computational pathology, but hospitals
typically restrict their analysis to isolated, air-gapped workstations because these gigapixel slides
contain highly sensitive patient data. On such systems the workflow for a single case is onerous:
i. Technicians copy the multi-gigabyte WSI to a removable medium and walk it to the secure
workstation
ii. The slide is partitioned into patches (≈7 min)
iii. Deep-learning inference (≈20 min)
With sequential processing and manual hand-offs, throughput stalls well below the 50 cases per
day target for routine diagnostics.
We present a privacy-preserving, cloud-enabled pipeline that removes the physical-transfer
bottleneck while maintaining strict confidentiality guarantees. The solution hinges on hardwarebased trusted execution environments (TEEs): TEE Encryptor (on-premises) and TEE Analyzer
(cloud).
Because computation now runs on elastic cloud hardware, multiple TEE Analyzer instances can
be launched in parallel. A deployment with ten enclaves cuts effective turnaround to minutes per
case and comfortably exceeds the 50-case-per-day target, all without exposing WSIs or
predictions in plaintext to the cloud operator. This solution offers order-of-magnitude
improvements in digital-pathology throughput while preserving patient privacy.
License type:
Publisher Copyright
Funding Info:
This research / project is supported by the National Research Foundation, Singapore and Infocomm Media Development Authority - Trust Tech Funding Initiative
Grant Reference no. : DTC-RGC-01
This research / project is supported by the National Research Foundation, Singapore - AI Singapore 100 Experiments Programme
Grant Reference no. : AISG2-100E-2023-108