Design specifications

[anton-sviridov]

Retina CLI is a command line client to use Retina — client for visual recognition of medical reports using artifical intelligence.

Usage:

• $ retina detect --email [email] --password [password] --image [image path] --database [local database path]
  1. Call a function login with parameters email (&str) and password (&str). If successfull, the function returns a token;
  2. Convert an image to webp;
  3. Call a function send_fundus with token (&str) and webp image in data url (&str). If successfull, get result (struct FundusData), if not — print an error message;
  4. Check if database with table exists. If not, initialize it first.
  5. Rename an image to hash name
  6. insert fundus data and hash name of image to table;
  7. Check if folder cache exists. If not, create it first;
  8. Copy an image with hash name in folder cache.

All flags are required.

Problems:

1. To create a unique name for an image
   • to link a name to it's ID in database
   • to insert default name → rename it in table as it's ID
   • to set UUID name to an image
   • to set hash name

MADR: set hash name, because it is useful to learn

• $ retina report --database [local database path]
  1. Check if database exists. If does, go on. If does not, print "database does not exists";

     Data exists locally. On different devices it is not the same, so command does not need to log in.

  2. Run "SELECT * FROM reports";

     Table "reports":

     • id
     • image: string
     • structure: FundusData
     • date: string

  3. Print query result in CLI

Output:

8743b52063cd84097a65d1633f5c74f5.jpg
1999-11-22T19:28:36Z
    exudates_in_macula: true,
    exudates_in_fovea: true,
    height: 120,
    width: 120,
    image: "data-url"
    macula: [1.0, 2.0, 3.0],
    hard_exudates: [[], [], []],
    intraretinal_hemorrhages: [[], [], []],
    soft_exudates: [[], [], []],
    fibrose: [[], [], []],
    laser: [[], [], []],
    microaneurysms: [[], [], []],
    neovascularization: [[], [], []],
    preretinal: [[], [], []],
    va: [[], [], []],

01dfae6e5d4d90d9892622325959afbe.jpg
2024-07-28T20:32:57Z
    exudates_in_macula: true,
    exudates_in_fovea: true,
    height: 120,
    width: 120,
    image: "data-url"
    macula: [1.0, 2.0, 3.0],
    hard_exudates: [[], [], []],
    intraretinal_hemorrhages: [[], [], []],
    soft_exudates: [[], [], []],
    fibrose: [[], [], []],
    laser: [[], [], []],
    microaneurysms: [[], [], []],
    neovascularization: [[], [], []],
    preretinal: [[], [], []],
    va: [[], [], []],

Stack:

• Rust
• Clap
• Serde
• Reqwest
• SQLite
• rusqlite
• Diesel

---

Sub issues:

• [x] #2
• [x] #3
• [x] #4
• [x] #5
• [x] #6
• [ ] #12
• [ ] #7
• [ ] #8
• [ ] #9
• [ ] #10

[anton-sviridov]

We need to print to user readable path to images. So we create a folder "cache", where give images unique names and add them to database table.

webp64 — data url i.e. it has file itself in url. The problem is that the string, returning webp, is too long for user

[anton-sviridov]

If we detect two same images, we will get two same outputs with possibly different dates/times

[anton-sviridov]

Feature request:

• Keep .sqlite file in local storage outside the project
• Keep cache folder in local storage outside the project. With optional flag --cache user can change path of cache folder

[anton-sviridov]

Divide functions and structures into modules: cli.rs, retina.rs