Automated Identification of Neoplasia in Diagnostic Imaging text reports

Transcription

1 Automated Identification of Neoplasia in Diagnostic Imaging text reports "This work has been funded in whole or in part with Federal funds from the National Cancer Institute, National Institutes of Health, Department of Health and Human Services, under Contract No. HHSN C George Cernile Artificial Intelligence in Medicine, Inc. Suzanne March QuantumMark LLC.

2 Objective To extend the use of natural language programming techniques to other sources of data beyond pathology reports.

3 Introduction The Team QuantumMark PI & Project Management AIM - Knowledge Engineering & Development April Fritz & Associates Domain Expertise, Cancer Registrar Participating Institutions 12 data providers including registries, labs and hospitals

4 Radiology Reports HMR MRI BRAIN W/WO CONTRAST - Aug FINDINGS: MRI of the brain with contrast. Clinical Indication: 82-year-old female with history of meningioma. Technique: MRI of the brain was obtained using the following sequences:sagittal T1, axial T1, axial FLAIR, axial T2, axial diffusion, axial ADC, axial postcontrast T1, and coronal postcontrast T1 weighted sequences. Comparison: comparison is made to an MRI of the brain from an outside institution (Downey Regional Med Ctr) dated 12/5/2006. There are no extra-axial fluid collections. A right frontal extra-axial parasagittal mass demonstrating isointense T1 signal to gray matter and intermediate to high signal intensity on T2 images with marked homogeneous enhancement postcontrast is noted to measure 2.6 cm transverse x 3.4 cm AP x 3.5 cm craniocaudal. Marrow signal changes in the calvarium abutting the mass are suspicious for interosseous involvement. When compared to the prior study of 12/5/2006, the size and appearance of the mass demonstrates no significant change. There is a minimal to mild mass effect on the adjacent right frontal lobe without evidence of edema. The cerebellum demonstrates mild atrophy. The brainstem appears normal. There are no parenchymal masses or midline shift. The are no restricted diffusion abnormalities. Mild right anterior ethmoid sinus mucosal disease is noted. A minimal amount of fluid is noted in the right mastoid air cells. The orbits, remaining paranasal sinuses, and calvarium are unremarkable. IMPRESSION: 1. Right frontal parasagittal meningioma, relatively unchanged in size and appearance when compared to the prior outside study dated 12/5/2006, including marrow signal changes in the calvarium adjacent to the lesion suggestive of intraosseous involvement. 2. Nonspecific white matter disease as described above, likely reflecting chronic ischemic changes. 3. Mild right anterior ethmoid sinus mucosal disease. Minimal amount of fluid noted in the right mastoid air cells.. The ventricles and cortical sulci are prominent consistent with mild to moderate cerebral atrophy. There is no hydrocephalus. The supratentorial brain parenchyma demonstrates scattered foci of T2 and FLAIR hyperintensities noted throughout the periventricular, subcortical, and deep white matter of both cerebral hemispheres, which are nonspecific, and likely represents chronic ischemic changes. The bilateral basal ganglia demonstrate T2 and FLAIR hypointensity.

5 Imaging terms 1 st pass Term Select Term Select hyperintensity FLAIR hyperintensity T2-weighted white matter change STIR hyperintensity T2 hyperintensity leukomalacic changes abnormal density white matter Yes choroid plexus cyst mass effect Maybe demyelinating disease. abnormal enhancement Maybe brain tumor Yes high density lesions heterogeneous enhancement Yes low density lesions Yes infarction / infarct abnormalities Maybe white matter signal abnormality abnormal enhancement chiari malformation small vessel disease chronic ischemic change extraaxial mass lesions Yes low attenuation changes Maybe increased signal intensity Maybe Edema / Oedema abnormal signal intensity Yes vasogenic Edema/Eodema Yes enhancement Yes BBB (Blood Brain Barrier) Yes pathologic mass Yes Infiltrative lesion Yes increased density Yes tumor Maybe retention cyst. Selection Rate > 30%

6 Imaging terms 2 nd pass Term Select Term Select hyperintensity FLAIR hyperintensity T2-weighted white matter change STIR hyperintensity T2 hyperintensity leukomalacic changes abnormal density white matter choroid plexus cyst mass effect Maybe demyelinating disease. abnormal enhancement brain tumor Yes high density lesions heterogeneous enhancement low density lesions infarction / infarct abnormalities white matter signal abnormality abnormal enhancement chiari malformation small vessel disease chronic ischemic change extraaxial mass lesions Yes low attenuation changes increased signal intensity Edema / Oedema abnormal signal intensity vasogenci Edema/Eodema enhancement BBB (Blood Brain Barrier) pathologic mass Yes Infiltrative lesion Yes increased density tumor Maybe retention cyst. Selection Rate ~ 10% 6

7 Process Clinical Indication Section % Selected with and without section filter No Clin Indication With Clin Indication Reno USC Desert

8 Phase I Results Selected 853 (8.7%) Selected 656 (7.3%) Selected TOTAL 4000 TOTAL, 9796 TOTAL, Selected, 144 (14.6%) 0 TOTAL, 983 Desert Radiology RENO Diagnostics USC

9 Phase II Installed into 3 registries and an initial run was done to determine performance and receive feedback. Different requirements One registry was very sensitive to false positives, and in fact wanted only tumors that were new to the registry. Another registry wanted all cancers for follow-up with existing registry data.

10 Two different case-findings 1. Identify reports of the diagnosis, differential diagnosis, metastasis or history of primary CNS neoplasms or non CNS neoplasms of behavior greater than Identification of a diagnosis or differential diagnosis of a primary CNS neoplasm.

11 Bundle Sensitivity and Specificity Results for QC Study 1289 Reports True Positive True Negative False Positive False Negative Sensitivity Specificity B % 97.8% B % 96.4% B % 95.1% B % 98.9% B % 97.6% B % 97.8% B % 93.9% B % 97.8% B % 96.8% B % 91.8% B % 98.9% B % 95.6% B % 96.8% Total % 96.6% 11

12 Extended Classifications Negative History of cancer Metastatic tumor Positive previously known Positive This information would allow increased decision making as well as provide information to registries.

13 Report classification Classification History of Metastatic CNS Non CNS 1 Negative History yes - - yes 3 Metastatic implied Yes implied implied 4 Positive previously known yes - yes - 5 Positive - - yes -

14 What about Sensitivity and Specificity? Classification History of Metastatic CNS Non CNS 1 Negative History yes - - yes 3 Metastatic implied Yes implied implied 4 Positive previously known yes - yes - 5 Positive - - yes -

15 What about Sensitivity and Specificity? Classification History of Metastatic CNS Non CNS 1 Negative History yes - - yes 3 Metastatic implied Yes implied implied 4 Positive previously known yes - yes - 5 Positive - - yes -

16 What about Sensitivity and Specificity? Classification History of Metastatic CNS Non CNS 1 Negative History yes - - yes 3 Metastatic implied Yes implied implied 4 Positive previously known yes - yes - 5 Positive - - yes -

17 What about Sensitivity and Specificity? Classification History of Metastatic CNS Non CNS 1 Negative History yes - - yes 3 Metastatic implied Yes implied implied 4 Positive previously known yes - yes - 5 Positive - - yes -

18 Positive Previously Known Resection medulloblastoma Post surgical change

19 Positive Pituitary gland microadenoma and no other neoplasm

20 History of cancer History of prostate cancer No other neoplasm

21 Metastatic History of renal cancer with lesions on frontal lobe

22 Negative 1 History of Metastatic Positive 5 CNS

23 Results Bundle True Positive True Negative False Positive False Negative Sensitivity Specificity B B B B B B Total Normal distribution Total Positive enriched Total Combined

24 Problem

25 What about sites beyond CNS? Attempted to process Pancreas and biliary reports. Not specific enough to isolate only to these organs Would have produced lots of reports that were already in registry, eg: liver biopsy, etc..

26 Tried to include Biliary, Pancreas reports problem!

27 Production system Standardized Format Standardized Format Dx Imaging Identification & Sending Receipt & Review Ca Registry Data HL7/XML Integration Coding and Filtering Secure IP Network HL7/XML Integration Data Review (optional) Data Siemens GE Hitachi Phillips Toshiba Fonar Etc. AIM Interfaceware Orion Mitem etransx Etc. 4 Impac Oncolog ERS Seer DMS CDC C/Next Etc. CNS Neoplasm Lexicon

28 Conclusion System implemented at three pilot sites QC studies will commence to determine Sensitivity/Specificity for selectable reports Determine accuracy in supplementary classifications System can be used as E-path add-on, or as standalone