Journal Article

DKFZ-2026-02473

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Explainable Artificial Intelligence Approaches for Autism Spectrum Disorder Classification via Hyperspectral Imaging Analysis of Red Blood Cells

Kalatzis, D. (First author)DKFZ* ; Spyratou, E. ; Vartian, R. ; Sansone, A. ; Parmeggiani, A. ; Grossi, E. ; Marini, M. ; Catania, V. D. ; Lima, M. ; Ghezzo, A. ; Panisi, C. ; Angotti, M. ; Abruzzo, P. M. ; Serpe, C. ; Efstathoupoulos, E. P. ; Ferreri, C. ; Kiouvrekis, Y.

2026
IEEE New York, NY

Abstract: Objective biological markers for Autism Spectrum Disorder (ASD) remain an importantresearch goal because clinical assessment is behavior-based and ASD is heterogeneous. This study evaluatedwhether eight red-blood-cell (RBC) hyperspectral endmember distributions contain participant-level ASDversus-neurotypical classification signal under leakage-controlled validation. The cohort comprised 58 malechildren (27 ASD and 31 neurotypically developing), each with nine RBC hyperspectral acquisitions(522 acquisitions in total). Six classifiers were evaluated using 58-fold outer leave-one-subject-outcross-validation, with five-fold participant-grouped inner hyperparameter selection. The nine held-outprobabilities were averaged to obtain one prediction per participant. XGBoost achieved the highest observedfixed-threshold accuracy (0.828; 95% CI 0.724–0.914), F1-score (0.800; 0.667–0.906), Matthews correlationcoefficient (0.657), and lowest Brier score (0.155). Support Vector Machine achieved the highest AUROC(0.879; 0.784–0.952), average precision (AP; 0.891), and specificity (0.968), but lower sensitivity (0.630).Exact paired McNemar tests found no significant participant-level accuracy difference between XGBoost andthe other models (all p ≥ 0.109). Out-of-fold TreeSHAP identified Spectrum 1, Spectrum 2, Spectrum 5, andSpectrum 4 as the leading XGBoost features. These model-specific associations are hypothesis-generatingand do not establish molecular or causal specificity. The findings support further investigation of RBChyperspectral profiles, but the small, male-only, single-cohort design and absence of external validationpreclude claims of clinical generalizability or diagnostic readiness.

Note: #EA:D120# / #NCTZFB26# / SCOPUS

---

Contributing Institute(s):

1. Angewandte Tumor-Immunität (D120)
2. Koordinierungsstelle NCT Heidelberg (HD02)

Research Program(s):

1. 314 - Immunologie und Krebs (POF4-314) (POF4-314)

Appears in the scientific report 2026

---

Database coverage:
; ; Article Processing Charges ; Clarivate Analytics Master Journal List ; Current Contents - Electronics and Telecommunications Collection ; Current Contents - Engineering, Computing and Technology ; DOAJ Seal ; Essential Science Indicators ; Fees ; IF < 5 ; JCR ; SCOPUS ; Science Citation Index Expanded ; Web of Science Core Collection

---