AHI POC

[PaulHax]

AHI browsing video:

aws-volview-demo-smaller.mp4

VolView load times with AHI and S3 Images

Results

Table compares VolView load times for DICOM series loaded from AHI and S3.

Cancer Imaging Archive Subject ID	Series UID	slices in series	AHI HTTP/2 load milliseconds	S3 HTTP/2 load milliseconds	AHI load milliseconds	S3 load milliseconds
129953	1.2.840.113654.2.55.157823367124358508737233300883051750223	57	2730		3258	3390
124227	1.2.840.113654.2.55.283562340021092685926251844637458324280	133	4457		6517	6220
214369	1.3.6.1.4.1.14519.5.2.1.7009.9004.759353208324955002724472043245	373	8216	12767	18512	19203

Images

3 CT series with 512x512 slices of varying numbers.

S3 files:
Served from S3 bucket via public http links. Original .DCM files used in the AHI dataset import.

AHI data:
Pulled from AHI api via Python FastAPI proxy.

Method

Run Python FastAPI proxy: floryst/aws-ahi-playground#1
Run VolView branch locally: #642

Put the URLs in the table in the browser.
Note console log output.

HTTP/2 Connection

For AHI FastAPI

Make certificate

sudo apt-get install openssl
openssl req -x509 -newkey rsa:4096 -keyout key.pem -out cert.pem -days 365 -nodes

pip install hypercorn
hypercorn src.aws_ahi.hello:app --bind 0.0.0.0:8000 --certfile ./cert.pem --keyfile ./key.pem

Start chrome without security

google-chrome --disable-web-security --user-data-dir="/somewhere" --ignore-certificate-errors

S3 HTTP/2 Setup

Configure CloudFlair for bucket

Conclusion

Factors for speedup of VolView’s DICOM image load times:

• Smaller data transfer of HTJ2K pixel data
• HTTP/2 connection

[PaulHax]

Just an experiment.