OrpheusAI

OrpheusAI is an intelligent diagnostic assistant designed to support dermatological analysis using deep learning. The system leverages a custom-built convolutional neural network (CNN) trained and tested on publicly available patient image data from DermNet.

The MVP implementation supports 10 disease classes, using a Keras-based CNN model optimized for image classification.

Architecture:

Backend: A Python server that loads the trained Keras model and serves predictions through a RESTful HTTP API. Frontend: A responsive client interface that allows users to upload images and receive real-time diagnostic insights.

The repository also includes a diseases.txt file that maps the model’s numerical outputs to human-readable class names, enhancing interpretability and ease of integration.

---

Repository structure

.
├─ .gitattributes
├─ README.md
├─ diseases.txt
├─ Backend/
│  ├─ Model.py
│  ├─ Model_v2_dev.py
│  ├─ requirements.txt
│  ├─ server.py
│  ├─ v1.keras
│  ├─ v2.keras
│  ├─ data/
│  └─ logs/
└─ Frontend/

Key components:

• Backend/server.py: Web server exposing model inference endpoints.
• Backend/Model.py: Model loading/inference utilities for the current model.
• Backend/Model_v2_dev.py: in-progress v2 model logic.
• Backend/v1.keras, Backend/v2.keras: Serialized Keras models.
• Backend/requirements.txt: Python dependencies for the backend.
• Backend/data/: Data artifacts used by the backend or training scripts (if any).
• Backend/logs/: Runtime or training logs (created by the backend during operation).
• diseases.txt: One label per line (e.g., disease/condition names) used to map model outputs.

---

How it works

1. Model assets

   • The backend uses a Keras model saved to disk (v1.keras or v2.keras).
   • diseases.txt contains the output label names (one per line). After inference, the backend maps raw model outputs (e.g., indices or probabilities) to these labels.

2. Backend server

   • server.py starts an HTTP API (commonly using Flask or FastAPI).
   • On startup, it loads the selected Keras model from Backend/ and prepares it for inference.
   • Incoming requests are parsed, data is preprocessed, inference is performed, and results are returned as JSON.
   • Logs may be written to Backend/logs/.

3. Frontend client

   • The Frontend/ directory holds the client UI (framework details depend on the files in that directory).
   • The frontend sends requests to the backend’s endpoints and displays model results.

---

Setup (Backend)

1. Create and activate a virtual environment:

   cd Backend
   python -m venv .venv
   # Windows:
   .venv\Scripts\activate
   # macOS/Linux:
   source .venv/bin/activate

2. Install dependencies:

   pip install --upgrade pip
   pip install -r requirements.txt

3. Ensure the model files exist:

   • v1.keras or v2.keras should be present in Backend/. (They are included in this repository.)
   • Ensure diseases.txt is at the project root and readable by the server.

---

Run (Backend)

There are two common ways to run the backend:

• Direct Python execution:

  cd Backend
  python server.py

---

Model development

• Model.py contains the primary model loading/inference utilities. If you’re customizing preprocessing or postprocessing, start here.
• Model_v2_dev.py is an experimental v2 version. Use it to iterate on new architectures or preprocessing flows.
• Training data and scripts:
  • If training utilities are included, they will likely read/write from Backend/data/ and log to Backend/logs/.
  • Adjust paths and parameters in the model scripts as needed.

Troubleshooting

• Missing dependencies: Reinstall with pip install -r Backend/requirements.txt.
• Model loading errors:
  • Verify MODEL_PATH and that the .keras file exists and is readable.
  • Ensure Keras/TensorFlow versions are compatible.
• Label mismatches:
  • Ensure diseases.txt label order matches the model’s output order.

---