Data leakage in feature selection invalidates reported classification accuracy for autism diagnosis from fMRI

EClinicalMedicine 2026 Open access

Read full text
Authors: Semen Kurkin, Alexander Hramov
DOI: 10.1016/j.eclinm.2026.104049
Cited by: 1
References: 3

Abstract

We read with interest the article by Vidya et al.1 reporting 98.2% accuracy for autism spectrum disorder (ASD) classification using deep learning on functional MRI data from the ABIDE I dataset. While the authors' focus on interpretability is commendable, we identified a critical methodological flaw that invalidates the reported performance metrics.

Fields of Study & Subjects

Autism Spectrum Disorder Research Face Recognition and Perception Emotion and Mood Recognition

Publication details

Journal / source
EClinicalMedicine
Publisher
Elsevier BV
Volume
97
Pages
104049–104049
Publication date
24 June 2026
Language
EN