Premium
Explorative analysis of IPA ‐ SPECT data through statistical inference for an automated diagnosis of glioma tumor
Author(s) -
Ubben Timm,
Kluge Andreas,
Abolmaali Nasreddin,
Iannilli Emilia
Publication year - 2018
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1002/mp.12770
Subject(s) - statistical parametric mapping , parametric statistics , statistical inference , computer science , workflow , glioma , statistical power , medical imaging , inference , artificial intelligence , spect imaging , data mining , pattern recognition (psychology) , nuclear medicine , magnetic resonance imaging , medicine , mathematics , radiology , statistics , cancer research , database
Purpose The identification of a brain tumor imaged with PET or SPECT is usually performed with visual inspection of an expert medical clinician. However an automated diagnostic of such images hasn't been established or applied. In this study, we explored the possibility of establishing an automated statistical analysis for the diagnosis of glioma by means of IPA ‐ SPECT data. Methods On the basis of a dataset of 100 patients that have undergone MRI and IPA ‐ SPECT acquisition, in this work, we identify an automated workflow. Three different approaches were explored: I. statistical non‐parametric mapping analysis (Sn PM ), II . statistical non‐parametric analysis with an increased number of permutations due to sign‐flipping function ( PALM ) and III . statistical parametric analysis ( SPM ). The automated methods were compared with the visual inspection. Results The study proved PALM and SPM approaches to have a high diagnostic power. Compared to the parametric methods, the non‐parametric method is the mathematically correct approach for the problem in question. If we take the high resolution structural MRI information into account, the diagnostic power of PALM was not significantly inferior to the visual inspection ( P = 0.5150), showing an area under the ROC curve ( AUC ) smaller only by less than 3%. Conclusions The automated diagnostic method based on statistical inference, here applied to diagnose glioma tumors in IPA ‐ SPECT data, seems to be a promising tool that can support the visual investigation in nuclear medicine. Moreover in the foreseeable future, the presented methodology has a big potential in various application like localization of active tumor tissues in surgical resection or stereotactic radiosurgery.