Radiomics and Deep Learning for Lung Cancer Screening
prostate RSNA2013v5edited
1. Characteristics of Prostate Cancer
Missed by Prostate MR: Correlation
with Histopathology
Nelly Tan, Daniel J. Margolis, David Y Lu, Robert E. Reiter, Steven S. Raman
Departments Radiology, Pathology & Urology
University of California, Los Angeles
2. Disclosures
• S. Raman & D. Margolis - Master Research
Agreement from Siemens medical systems
• UCLA SPORE
3. Introduction
• MR Imaging of the prostate may help guide
clinicians to stratify patients to appropriate
treatment
– Active surveillance an alternative to definitive
therapy
– Emerging ablative local therapy
• Understanding of the features of prostate
tumors missed by 3T multi-parametric
prostate MRI important for targeted biopsy.
4. Objective
• To determine the characteristics of prostate
cancer foci missed by mulit-parametric MRI.
5. Methods
• A HIPPA-compliant, IRB-approved retrospective study of 122
patients
• Inclusion Criteria
– Underwent MRI using an endorectal coil (eMRI)
• T2 Fast Spin Echo, Diffusion weighted imaging, Dynamic contrast-
enhanced T1 weighted MR, MR Spectroscopy
– Subsequently had prostatectomy
• Exclusion
– MR performed at outside facility (n=3)
6. Methods
• Prospective evaluation of pathologic features by
uropathologist blinded to MR used as gold
standard
– Gleason Score, size of tumor
– Pathological stage: ECE, SVI
– Location: Base, Mid, Apex
– Index lesion (vs satellite): largest or highest Gleason
score
7. Methods
• Retrospective independent review by
uroradiologist and a second pathologist
collectively matched each tumor.
• Recorded on a prostate template
• MRI :
– Any lesion worrisome on any of the MR
parameters was considered an MR lesion
8. Methods
• Standardized classification system was used to
characterize the multi-parametric MR features
– based on Likert scale (1-5)
• Statistical analysis:
– Chi-square analysis was performed for categorical
and t-test for continuous variables.
– Multivarible logistic regression to identify
predictors for tumor detection.
9. Results
• 122 patients had 176 unique suspicious lesion
called by MR
• 149 (52.5%) prostate tumors in 74 patients
were missed by MRI.
10. Results
Variables
Patients, n 122
PSA ng/cc, mean (SD) 7.2 (5.9)
Total tumors, n 285
Tumor diameter, mean cm(SD) 1.3 (0.99)
Focality
Single tumor
≥2 tumors
26 (21.3)
96 (78.7)
Total tumors, 285
GS6, n(%)
GS3+4, n(%)
GS4+3, n(%)
GS 8-10, n(%)
151 (53.0)
77 (27.0)
38 (13.3)
19 (6.7)
Location, n=285
Apex, n
Mid/Base, n
59 (20.7)
226 (79.3)
Stage
T2, n (%)
T3a, n (%)
T3b, n (%)
87(71.3)
30 (24.6)
5 (4.1)
11. Pathologic
examination,
tumor present
Pathologic
examination,
tumor absent
Total P-value
MRI "called" Lesions, Total n
Yes
No
285
135 (47.3)
150 (52.6)
45
45 (25.0)
0
330
180
150
<0.01
Detection of Index tumor
Index lesion
Satellite tumors
100 (82.0)
35
22
128 (78.5)
122
163
<0.01
PSA, mean ng/ml (SD) 7.7 (7.2) 7.1 (5.4) 7.2 (5.9) 0.44
Tumor diameter
<1 cm, n (%)
≥ 1 cm, n (%)
26 (19.8)
109(70.8)
105(80.1)
45 (29.2)
131
154
<0.001
Total tumors, n =285
GS6, n (%)
GS3+4, n (%)
GS4+3, n (%)
GS 8-10, n (%)
39 (25.8)
54 (70.1)
29 (76.3)
13 (68.2)
112 (74.2)
23 (29.9)
9 (23.4)
6 (31.6)
151 (53.0)
77 (27.0)
38 (13.3)
19 (6.7)
<0.01
Location, n=285
Apex, n (%)
Mid/Base, n (%)
13 (22.0)
118(52.2)
46 (77.9)
102 (45.1)
59 (20.7)
226 (79.3)
<0.01
PPV 75%
12. Results
• Simple and multivariable logistic regression
were performed to identify predictors of
prostate cancer detection.
Prostate cancer detection Crude OR
(95% CI)
Adjusted OR
(95% CI)
Prostate cancer detection:
Index (vs. satellite)
Apical lesion
Tumor diameter, cm
Gleason ≥3+4
16.62 (9.17-30.11)*
0.244 (0.12-0.48) *
5.17 (3.42-7.82) *
7.25 (4.30-12.24) *
6.62 (3.06-14.31) *
0.36 (0.16-0.83) *
2.24 (1.36-3.70) *
1.31 (0.62-2.74)
16. 67 yo M, PSA 4.6 6
Index GS 3+4 focus detected
Apical GS 3+3(5mm) missed
GS 3+3 missed
17. Summary
• Prostate cancer foci missed on MR were
– Smaller in maximal diameter
– Higher in proportion of low-grade tumors (GS6)
– Commonly located in the apex
• After adjusting for significant variables, apex,
size and index lesions are most important.
• MR imaging may help guide patient selection
• Lesions the apex are difficult to identify
18. Limitations
• Retrospective study
• Single institution, single surgeon
• Although all cases were scored prospectively,
and scoring system has been validated (Sonn
et al, European urology 2013), it does not
adhere to the ESUR PI-RADS specifications