About this skill

Problem

45-degree fundus color photo AI analysis is not just one synchronous API call. It requires converting JPEG, PNG, or DICOM images to raw Base64, deriving left/right positions for images[].descPosition, uploading with a self-generated studyId, polling the asynchronous result, and filtering valid status = 200 entries from glaucomaResultList and multipleDiseasesResultList. It also involves handling asynchronous status codes, extracting report links, and managing left/right encoding differences across fields.

How It Works

The Skill wraps the Tencent Health Miying fundus AI HTTP flow:
- Upload the study: call studyupload/v2/{appId} with fixed studyName, studyType=2, and 1 to 20 images;
- Authenticate requests: send appId and timestamp headers, and generate signature using HMAC-SHA256(token, appId + timestamp);
- Poll results: call queryEyeAIResult/{appId}, using aiType=0 for glaucoma plus multi-disease analysis and needReport=1 for a PDF report; polling waits at 10-second intervals, code=30008 means processing, and code=0 with non-empty data means success;
- Format output: mark normal, abnormal, and unknown findings with ✅/⚠️/❓, extract reportUrl, and summarize the diagnosis.

Boundaries and Notes

Left/right encoding differs by field: upload descPosition uses 1=left, 2=right, while result eyeCategory uses 0=left, 1=right. Filenames containing OD, _R, or right are inferred as right eye; OS, _L, or left as left eye; otherwise unknown is used. Do not include the data:image/...;base64, prefix in Base64 payloads; keep single files around 5MB or less; use about 60 seconds for upload timeouts and 30 seconds for query timeouts; keep each studyId unique. It suits engineering pipelines that need fundus image onboarding, async status handling, and structured result formatting. Trial credentials have limited calls and concurrency, so production use requires separate authorization.

Screenshots

Use Cases

{'zh': 'Convert JPEG, PNG, or DICOM fundus images to raw Base64, generate `images[].descPosition`, and upload the study', 'en': 'Convert JPEG, PNG, or DICOM fundus images to raw Base64, generate `images[].descPosition`, and upload the study'}
{'zh': 'Upload fundus photos with a self-generated `studyId`, then poll `queryEyeAIResult` at 10-second intervals until a valid result is returned', 'en': 'Upload fundus photos with a self-generated `studyId`, then poll `queryEyeAIResult` at 10-second intervals until a valid result is returned'}
{'zh': 'Filter `status = 200` entries from `glaucomaResultList` and `multipleDiseasesResultList`, then extract `reportUrl`', 'en': 'Filter `status = 200` entries from `glaucomaResultList` and `multipleDiseasesResultList`, then extract `reportUrl`'}
{'zh': 'Handle left/right encoding differences in result parsing to avoid confusing upload `descPosition` with result `eyeCategory`', 'en': 'Handle left/right encoding differences in result parsing to avoid confusing upload `descPosition` with result `eyeCategory`'}

Best For

{'zh': 'Healthcare software engineers who need to integrate 45-degree fundus photos into an AI diagnosis HTTP pipeline', 'en': 'Healthcare software engineers who need to integrate 45-degree fundus photos into an AI diagnosis HTTP pipeline'}
{'zh': 'Backend developers who need to handle Base64 image upload, HMAC-SHA256 signing, and asynchronous polling', 'en': 'Backend developers who need to handle Base64 image upload, HMAC-SHA256 signing, and asynchronous polling'}
{'zh': 'Engineers who need to format fundus AI results into structured output with status icons and report links', 'en': 'Engineers who need to format fundus AI results into structured output with status icons and report links'}
{'zh': 'Integration engineers who need to distinguish upload eye-position fields from result eye-category fields to avoid left/right parsing errors', 'en': 'Integration engineers who need to distinguish upload eye-position fields from result eye-category fields to avoid left/right parsing errors'}

Related Skills

Cold Shower Gate

@user_2da54b45/cold-shower

Switch AI from a polite executor into a skeptical thinking partner, using assumption checks, pre-mortems, and evidence gates to expose weak ideas early.

Professional 2026.08.29 ↓ 280