Radiomics: bringing radiology into the 21 st century Dr Tim Rosenow
How radiology operates
Sometimes hilariously vague • Normal abdomen radiographically with no visualized acute diagnostic abnormalities evident within the abdomen on this examination at the present time radiographically. Opinion: Abdomen within the range of normal
Reports are poorly repeatable • “in patients with pneumonia, the interpretation of the chest X-ray, especially the smallest of details, depends solely on the reader.” Moncada 2010 Braz J Infect Dis
Difficult to compare over time • Baseline scan: – “Numerous clusters of small nodules, likely inflammatory in nature.” • Six months later: – “Heterogeneously distributed clusters of inflammatory nodules.”
Radiomics • Quantitative data extracted from medical images • Small or no involvement by humans – Unbiased, objective, sensitive – Less labour intensive
Simple analytics
Example: cardiac disease
Example: lung V’ and Q’
3D models
3D model simulations http://vasclab.mech.uwa.edu.au
4D data analysis
Artificial intelligence pipelines PRAGMA-CF: a case study
Early intervention USCF Data Registry Annual Report 2015
AREST CF Early Surveillance Program Birth 3 months 1 year Annually to 5 years Clinical progress Bronchoalveolar lavage Chest CT scan Epithelial samples Lung function (MBW, RVRTC, FOT) Exhaled Breath Condensate QoL/Psychosocial
CT lung disease
Morphometric analysis Legend: 1. “Normal” lung 2. Bronchiectasis 3. Mucous plugging / consolidation 4. Bronchial wall thickening 5. Atelectasis
Biologic validation Neutrophils* N. Elastase + PRAGMA-CF 0.41 (0.025) 0.004 CT score 0.31 (0.096) 0.110 * Spearman’s rho (P-value) + Wilcoxon rank-sum P-value N = 60 scans N = 683 scans
Databank and training • Requires collaboration Machine 300 + 900 patients learning 1000 + 2500 scans 35,000 annotated slices 3.5 million annotations
Validation and application • Reserve dataset • Interface, integration and regulation OR – FDA • New dataset – EMA • Testing, tweaking – TGA
Radiologist workflow
Generic pipeline
Radiologists of the future
Recommend
More recommend