Psychiatry · Hum Brain Mapp · 2022
The paper’s best brain measure for spotting schizophrenia: 84.09% accurate. Its table: 80.68%; another beats it.
Spotting schizophrenia: share of people labelled correctly (coin-flip guessing gets 50%)
The paper says
The paper rates how irregular brain activity is in scans with seven measures, and has a computer use each to tell healthy people from people with schizophrenia, bipolar disorder or ADHD. For schizophrenia it names one measure, “range entropy”, best, labelling 84.09% of people correctly.
Its own data show
Its own results table gives 21 schizophrenia results: each measure on three ways of dividing the brain into regions. The top is “sample entropy”, 82.95%; range entropy’s best is 80.68%, second. None reaches 84.09%. That number is in the table, but as a different score: range entropy called 84.09% of the healthy people healthy.
Accuracy counts everyone: patients labelled ill plus healthy people labelled healthy; coin-flip guessing gets 50%. A score on healthy people alone ignores the patients, so it is not accuracy. The summary’s main message is which measure suits each illness. For schizophrenia, its own table points to sample entropy, not range entropy. Its ADHD and bipolar picks match the table.