The idea is to have a complete system for acquisition, detection and report of malaria cases
Acquisition is carried out by blood smear stained with Giemsa (protocol below) and visualized on a microscope with USB camera. Captured images and then processed and feeded to a pretrained DeepLearning architecture. Ultimately the diagnosis is mapped on a GIS system by using ArcPy. The deeplearning architecture should run on a normal PC, a RaspberryPI or be integrated into an Android app.
Giemsa Staining (thin film protocol):
- On a clean dry microscopic glass slide, make a thin film of the specimen (blood) and leave to air dry.
- Dip the smear (2-3 dips) into pure methanol for fixation of the smear, leave to air dry for 30seconds
- Flood the slide with the Giemsa stain solution.
- Flush with tap water and leave to dry Different solution and time with Giemsa stain solution can be used, [check this report] (https://journals.sagepub.com/doi/full/10.1258/td.2010.100218?casa_token=8rA9Ezu6CvAAAAAA:313ryHWFaL0Vbv5aWVptYuDoTZbwm-RwrkZobBYeJXn6DmkMn8OfjxqmNO5IbyBKmK9exSdxkkfEDQ)
GoogleColab with VGG19 without finetuning: https://colab.research.google.com/drive/1CUcsmtg9S9ryJZAZrQsAf7e5ouUK4gL8#scrollTo=-NJ2CxM2Rf5d
to do:
- Testing other blood samples
- Retrain thick smear samples
- Android app
- Connection to ArcGIS server (consider migrating to AWS by using GeoDjango https://realpython.com/location-based-app-with-geodjango-tutorial/ )
- Write cell extraction using SK-image
