Staged reflexive artificial intelligence driven testing algorithms for early diagnosis of pituitary disorders

Area of Science:

• Endocrinology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Sellar masses often cause subtle hormonal dysfunction.
• Manual laboratory clinician review of endocrine tests can identify pituitary disorders but is not scalable.
• Previous methods involved a combined reflex/reflective approach for diagnosing pituitary dysfunction.

Purpose of the Study:

• To compare the accuracy and financial impact of an AI-based computerized reflex protocol versus a manual protocol.
• To evaluate the efficacy of automated reasoning in multi-stage reflex testing for pituitary dysfunction.

Main Methods:

• Developed a proof-of-concept AI framework for automated multi-stage reflex testing.
• Compared AI protocol performance against manually curated retrospective data over 12 months.
• Utilized automated reasoning methods for pituitary dysfunction detection.

Main Results:

• The AI protocol matched or included all manual add-on tests in 92% of cases.
• AI recommended similar specialist referrals in 90% of cases.
• AI issued 2.8x more lab tests at 85% lower operational cost than the manual protocol.

Conclusions:

• AI-based reflex protocol successfully identifies pituitary dysfunction with reduced laboratory costs.
• Future work will enhance AI accuracy and integrate it into healthcare systems.
• This automated approach facilitates wider screening for undiagnosed pituitary disorders.