- Standardized protocols should be compiled so to collect reproducible data. Actually these protocols are not present and the laboratories operate with different approaches.
- The biological model is quite complex. Apply machine learning to correctly classify a disease it is not so easy as to discriminate the presence of a mineral extracted from the soils. Human exhibits a compless biochemical equilibrium and the data to be elaborated should be carefully acquired and selected to obtain the correct diagnosis.
- The correct information should be inserted in the machine learning database to increase its discrimination power. It is not sufficient to provide data of an analytical instruments and basing all on a kind of instrumental signal. Even the anamnesis data of the patients like age, life stile etc can add specificity and selectivity to the data elaboration. These data should be collected by certified group so to avoid mistake during the database data entry process.
Machine learning is surely a good opportunity to help clinic to make diagnosis and found therapy but the follow activities are expected in the future to increase its applicability:
- Define the type of data to be collected.
- Define standard protocols for data collection.
- Optimize the machine learning algorithm on the complex biological model.
- Validate the machine learning technology by means of inter-laboratory studies in certified institutes. The diagnostic performance should be clearly expressed in term of selectivity, specificity and sensitivity.
We will follow the development of this interesting field and we will keep in touch to share the news.