Tkinter CNN Alzheimer Detection App
Budget: $8 – $15 USD
I want to package my existing deep-learning work into an easy-to-use desktop tool built with Python + Tkinter. The goal is simple: a clinician or researcher drops a PNG brain scan into the interface, presses “Analyse”, and instantly receives a prediction on whether the image shows signs of Alzheimer’s disease.
Here is what I need implemented:
• A clean Tkinter GUI with buttons to load single or multiple PNG images, display a thumbnail preview, and trigger inference.
• Integration of a Convolutional Neural Network model (Keras / TensorFlow preferred). You may start from a proven architecture such as VGG, ResNet or a lightweight custom CNN—what matters is solid accuracy and clear code.
• On-screen prediction output (probability score and “Alzheimer-positive / negative” verdict) plus an optional CSV export for batch runs.
• Error handling for non-PNG inputs and helpful status messages while the model loads or processes images.
• A short README explaining environment setup, model training pipeline, and how to retrain on new datasets.
Acceptance criteria
1. Running `python app.py` opens the GUI and processes sample PNG images I will supply.
2. Average inference time per image ≤ 2 seconds on a CPU-only machine.
3. Prediction accuracy meets or exceeds the benchmark I will share (based on my validation set).
If you can also add optional JPEG/DICOM support later, let me know—future flexibility is a plus, but right now we only target PNG brain scans. Clean, well-commented code and clear instructions will be key to successful delivery.
Here is what I need implemented:
• A clean Tkinter GUI with buttons to load single or multiple PNG images, display a thumbnail preview, and trigger inference.
• Integration of a Convolutional Neural Network model (Keras / TensorFlow preferred). You may start from a proven architecture such as VGG, ResNet or a lightweight custom CNN—what matters is solid accuracy and clear code.
• On-screen prediction output (probability score and “Alzheimer-positive / negative” verdict) plus an optional CSV export for batch runs.
• Error handling for non-PNG inputs and helpful status messages while the model loads or processes images.
• A short README explaining environment setup, model training pipeline, and how to retrain on new datasets.
Acceptance criteria
1. Running `python app.py` opens the GUI and processes sample PNG images I will supply.
2. Average inference time per image ≤ 2 seconds on a CPU-only machine.
3. Prediction accuracy meets or exceeds the benchmark I will share (based on my validation set).
If you can also add optional JPEG/DICOM support later, let me know—future flexibility is a plus, but right now we only target PNG brain scans. Clean, well-commented code and clear instructions will be key to successful delivery.