ARTIFICIAL INTELLIGENCE
( continued from page 11 ) vital for accurate diagnosis .
New computerized techniques for measurement of bone age are using machine learning to become much more accurate and precise , can be used for both short and tall children , and are adaptable to various ethnicities . The BoneXpert system , which uses machine learning , though not deep-learning , is the only medically certified and systematically validated technique . The techniques are still developing , with mistakes occasionally occurring , and at present syndromes and anomalies cannot be automatically detected . Therefore , clinical experience and judgment is still required .
A variety of models to predict the response to growth hormone ( GH ) treatment in individual children with growth failure have been developed and validated over time . However , they are still largely underused and require clarity in the models being used , expert validation and assessment by endocrinologists .
Some prediction models use IGF-1 SDS ( Insulin-like growth factor 1 standard deviation scores ) as a factor , and IGF-1 SDS may also be used to evaluate response to GH treatment for short stature . Machine learning techniques indicated that baseline IGF-1 SDS is among the most important indicators of response to GH treatment . However , IGF-1 measurements should be considered with care because results can vary widely between different assays , the normative data used to determine SDS values , and pubertal stage of the child in addition to age , gender and nutritional history , when interpreting results . Titrating GH dose to IGF-1 helps to take into account the sensitivity of treatment due to diagnostic factors and has been reported to provide better clinical outcomes .
2 ) Facilitating diagnostic workflow by tackling gray areas of diagnostic uncertainty
Asymptomatic hyperparathyroidism can be challenging to identify without a high index of suspicion because it involves subtle biochemical changes and its phenotype overlaps with those of primary osteoporosis and other rare mineral disorders , including familial hypocalciuric hypercalcemia . Training ML models with datasets obtained from patients may help resolve these grey areas . 5
Several studies have shown that ML could support the decision process of whether to perform an invasive biopsy on a thyroid nodule based on ultrasonography , with good classification performance similar to that of radiology experts ; therefore , ML classifications might potentially provide guidance to operators during data acquisition and measurement . 5
A well-validated , accurate , non-invasive ML model may have the potential to replace standard invasive diagnostic modalities for certain diseases . For instance , the global burden of nonalcoholic fatty liver disease ( NAFLD ) is rapidly growing , but invasive liver biopsy remains the gold standard for diagnosing NAFLD and nonalcoholic steatohepatitis . Perakakis et al . developed a support vector machine-based model to classify NAFLD based on features obtained from the lipidomic , glycomic , and liver fatty acid analysis
12 LOUISVILLE MEDICINE of serum samples . For liver fibrosis , an exploratory model with 10 lipid species showed high accuracy ( up to 98 %), suggesting the possibility of a targeted lipidomic approach as an alternative non-invasive diagnostic tool , although the model needs to be further validated in other ethnicities and individuals with a milder spectrum of liver diseases . 5 , 6
3 ) Finding novel disease clusters and associations using patient databases
4 ) Risk prediction including clinical outcomes and treatment response
Clinical Outcomes : Accurately predicting clinical outcomes enables an individualized approach to treatment strategy and monitoring . The Weight , Age , hyperTension , Creatinine , High-density lipoprotein cholesterol , Diabetes control , and Myocardial infarction ( WATCH-DM ) score was developed to predict heart failure risk among patients with Type 2 diabetes using ML algorithms based on the Action to Control Cardiovascular Risk in Diabetes ( ACCORD ) trial dataset , and showed good predictive performance with an external validation set ( the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial [ ALLHAT ]). 5
Treatment responses : Well-trained ML models may have the potential to provide guidance for dose adjustment , particularly in patients with chronic conditions requiring the indefinite replacement of certain hormones , as in patients who receive thyroid hormone replacement after total thyroidectomy or in Type 1 diabetes patients who receive insulin replacement .
AI in Type 1 diabetes management : People who manage their glucose levels using either continuous subcutaneous insulin infusion ( CSII ) or multiple daily injection ( MDI ) therapy must navigate a complicated landscape of “ hands on ” interactive guidelines for the maintenance of basal insulin doses and meal and correction bolus doses . This can be challenging because there are a number of therapy parameters that impact insulin dosing : pre-prandial glucose level , the grams of carbohydrate that they will consume , their insulin sensitivity , their specific insulin-to-carbohydrate ratio , and the current insulin-on-board ( IOB ). People may also need to consider insulin variations that can occur throughout the day , their current glucose trend , and the activity context under which an insulin dose is being taken ( e . g ., prior to exercise , during an illness , etc .). This is particularly difficult for people using MDI therapy , as compared to a person using a pump with a bolus calculator , although more recent smart insulin pens have recently made bolus calculation possible for MDI users . Some users may lack the arithmetic skills necessary to accurately calculate their insulin prior to meals and throughout the day . 7 , 8 , 9
A large number of users miscalculate their post prandial insulin need , often resulting in repeated hypoglycemia and hyperglycemia . The increasing ubiquity of health-based mobile computing and the growing usage of wireless continuous glucose monitors ( CGM ) have created an opportunity for development of automated decision support systems ( DSSs ) for people with Type 1 diabetes . While some