Artificial intelligence for detection of intracranial haemorrhage on head computed tomography scans: diagnostic accuracy in Hong Kong

INTRODUCTION

The use of artificial intelligence (AI) to identify acute intracranial haemorrhage (ICH) on computed tomography (CT) scans may facilitate initial imaging interpretation in the accident and emergency department. However, AI model construction requires a large amount of annotated data for training, and validation with real-world data has been limited. We developed an algorithm using an open-access dataset of CT slices, then assessed its utility in clinical practice by validating its performance on CT scans from our institution.

METHODS

Using a publicly available international dataset of >750 000 expert-labelled CT slices, we developed an AI model which determines ICH probability for each CT scan and nominates five potential ICH-positive CT slices for review. We validated the model using retrospective data from 1372 non-contrast head CT scans (84 [6.1%] with ICH) collected at our institution.

RESULTS

The model achieved an area under the curve of 0.842 (95% confidence interval=0.791-0.894; P<0.001) for scan-based detection of ICH. A pre-specified probability threshold of ≥50% for the presence of ICH yielded 78.6% accuracy, 73% sensitivity, 79% specificity, 18.6% positive predictive value, and 97.8% negative predictive value. There were 62 true-positive scans and 22 false-negative scans, which could be reduced to six false-negative scans by manual review of model-nominated CT slices.

CONCLUSION

Our model exhibited good accuracy in the CT scan-based detection of ICH, considering the low prevalence of ICH in Hong Kong. Model refinement to allow direct localisation of ICH will facilitate the use of AI solutions in clinical practice.

Immunoglobulin G4-related disease in Hong Kong: clinical features, treatment practices, and its association with multisystem disease

INTRODUCTION

Immunoglobulin G4-related disease remains an under-recognised and evolving disease. Local data are sparse and previous publications have been limited to individual case reports or case series only. We conducted this study to review the clinical features, treatment practices, and factors associated with multisystem involvement in Hong Kong. We described the clinical features and treatment modalities of the largest cohort of immunoglobulin G4-related disease in our locality thus far.

METHODS

We retrospectively evaluated all patients with immunoglobulin G4-related disease between January 2003 and December 2015 in Queen Mary Hospital and combined this with patient data extracted from previous local publications. We analysed the clinical features, treatment practices, and factors associated with the number of organ systems involved.

RESULTS

A total of 104 patients (55 from Queen Mary Hospital and 49 from literature review) were identified. Patients were predominantly older men (mean [standard deviation] age, 61.9 [12.7] years; male-to-female ratio=3:1) and 94.4% had elevated pre-treatment serum immunoglobulin G4 levels. Hepatobiliary and pancreatic system (40.4%), salivary gland (33.7%), lymph node (29.8%), and eye (19.2%) were the most common organ systems involved. Lymphadenopathy was associated with glucocorticoid use (odds ratio=2.65; 95% confidence interval, 1.08-6.54; P=0.034). Pre-treatment serum immunoglobulin G4 levels correlated with the number of organ systems involved (β=0.347; P=0.004) and, specifically, more associated with patients having salivary gland involvement than those without (mean, 1109 mg/dL vs 599 mg/dL; P=0.012).

CONCLUSION

We identified pre-treatment serum immunoglobulin G4 to be associated with multisystem disease, especially with salivary gland involvement, highlighting its potential for disease prognostication and monitoring. Increased physician awareness and multidisciplinary efforts are required for early diagnosis and optimal management of this masquerading disease.

Abstract P3-08-14: Could the preoperative systemic therapy be a risk factor for breast cancer-related lymphedema in stage II/III breast cancer?

Background: The breast cancer-related lymphedema (LE) has been known to be closely related to axillary lymph nodes dissection (ALND), chemotherapy, and radiation therapy. In this study, we evaluated whether the sequence of systemic chemotherapy and surgery could be a predictive factor in stage II/III breast cancer.

Methods and Materials: A total of 867 patients with stage II/III breast cancer, who underwent curative surgery with adequate systemic therapy from 2004 to 2009, were retrospectively analyzed. Adjuvant chemotherapy (ACT) was performed in 571 patients (65.9%) and preoperative systemic chemotherapy (PSC) in 296 (34.1%). We evaluated the incidence of LE by clinicopathologic factors and treatments.

Results: At a median follow-up of 5.1 years (range, 3.0-8.3 years), 360 patients (41.5%) had experienced LE, 244 patients have retained LE (permanent LE), and 116 patents were normalized. The overall 5-year cumulative incidence of LE was 17%. LE occurred in 188 patients (32.9%) in patients with ACT, 172 patients (58.1%) with PSC (P&lt;0.001), permanent LE in 121 (21.2%) with ACT, 123 (41.6%) with PSC (P&lt;0.001), respectively. Multivariate analysis showed that PSC (hazard ratio [HR], 1.65; P&lt;.001), radiotherapy (HR, 2.24; P&lt;0.01), ALND (HR, 1.41; P = 0.04), and nodal stage (HR, 1.93; P = 0.04) were independent risk factors for LE occurrence. For the permanent LE, PSC (HR, 1.44; P = 0.05), radiotherapy (HR, 2.79; P&lt;0.01), ALND (HR, 1.77; P&lt;0.01), and nodal stage (HR, 3.01; P = 0.02) showed the associations.

Conclusions: The risk factors associated with LE were advanced stage, ALND and radiotherapy. PSC was one of predictors for transients LE. However, further evaluation should be done whether it is a risk factor for permanent LE.

This research was supported by National Cancer Center Grant NCC-1210181-2 by the National Cancer Center, Republic of Korea.

Citation Information: Cancer Res 2013;73(24 Suppl): Abstract nr P3-08-14.

Abstract P4-09-14: Invasive Lobular Carcinoma Is a Prototype of Luminal A Breast Cancer Subtype and Rare in Korea

Purpose Invasive lobular carcinoma (ILC) is known to be the second most frequent histologic subtype, occupying 10% of invasive breast cancer in the Western countries. The present study was designed to assess the clinical characteristics and outcomes of ILC compared to general invasive ductal carcinoma (IDC) and the luminal A subtype (LA-IDC).

Methods The study population included d 2916 patients with invasive breast cancer consecutively diagnosed at the National Cancer Center, Korea between 2001 and 2008. The clinicopathological characteristics and clinical outcomes were retrospectively reviewed.

Results There were 83 pts (2.8%) diagnosed with ILC and 1,088 pts (37.3%) with LA-IDC. Mean age was 48.2 years of all patients, 48.3 years of ILC group and 47.9 years of LA-IDC group. The ILC patients presented with a larger tumor size (≥T2, 59.8% vs. 38.8%, P=0.001), lower histologic grade (HG 1 or 2, 90.4% vs 64.4%, P&lt;0.001), more often estrogen receptor (ER) and progesterone receptor (PgR) positivity (ER+, 90.4% vs. 64.4%, P&lt;0.001; PgR+, 71.1% vs. 50.1%, P&lt;0.001), HER2 negativity (97.5% vs. 74.9%, P&lt;0.001), lower Ki-67 expression (10.3% ± 10.6% vs. 20.6% ± 19.8%, P&lt;0.001), and luminal A subtypes (91.4% vs. 51.2%, P&lt;0.001) compared to the IDC group. Six (7.2%) ILC patients and 359 (12.7%) IDC patients developed disease recurrence with a median follow-up of 56.4 months (range 4.9-136.6 months). Although ILC showed similar prognosis to IDC in general (5-year DFS rate, 91.7% in ILC vs. 87.4% in IDC, P=0.31; 5-year OS rate, 93.6% in ILC vs. 92.5% in IDC, P=0.38), its outcome was closer to LA-IDC, and better than non LA-IDC (LA-IDC (ref); ILC, HR 0.77 in recurrence, 95% CI 0.31-1.90, P=0.57; HR 0.75 in death, 95% CI 0.18-3.09, P=0.70; non LA-IDC, HR 1.69 in recurrence, 95% CI 1.23-2.33, P=0.001; HR 1.50 in death, 95% CI 0.97-2.33, P=0.07) in univariate and multivariate analysis.

Conclusions ILC is a very rare histologic subtype of breast cancer in Korea compared to the Western countries and has distinctive clinicopathological characteristics similar to those of LA-IDC. Acknowledgement: supported by NCC grant #0910320

Citation Information: Cancer Res 2010;70(24 Suppl):Abstract nr P4-09-14.

Abstract P2-09-01: Serial [18F] FDG-PET after the 2nd Cycle of Preoperative Chemotherapy Is Predictive for Pathological Complete Response in Stage II/III Breast Cancer

Purpose: One of substudies of the prospective trials aimed to evaluate the usefulness of serial [18F] 2-fluoro-2-deoxy-D-glucose-positron emission tomography ([18F] FDG-PET) for predicting pathological complete response (pCR) in stage II/III breast cancer with preoperative chemotherapy (PST).

Methods: Serial PET was undertaken in 57 breast cancer patients enrolled in three different neoadjuvant trials: 35 patients from a phase II study with paclitaxel/gemcitabine/trastuzumab with ClinicalTrial.gov NCT 00532857, 9 patients from a phase Ib study with paclitaxel/gemcitabine/lapatinib with ClinicalTrial.gov NCT 01133912, and 13 patients from a phase Ib with paclitaxel/gemcitabine/sunitinib with ClinicalTrial.gov NCT0 1070706. All patients received 6 cycles of PST followed by surgery and radiotherapy. We assessed the peak standardized uptake value (SUVp) in the primary tumor at the baseline and after the 2nd cycle (37 patients) or after completion (20 patients) of 6 cycles of PST, and calculated the reduction rate (RR) of the SUVp. Pathological response was classified into pCR and non-pCR. To compare the mean of SUVp and RR of SUVp between different response groups, two-way tables and chi-square tests were used

Results: Fifteen (40.6%) of 37 patients who took repeat PET after the 2nd PST and 15 (75%) of 20 patients after completion of PST achieved a pCR with overall pCR rate of 52.6% in the primary tumor. In patients with repeat PET after the 2nd PST, post-treatment SUVp and RR of the SUVp in primary tumors were significantly different by the pathological response (post-treatment SUVp, 1.54 ± 0.63 in pCR vs 2.54 ± 1.06 in non-pCR, P=0.002; RR of the SUVp, 79.2% ± 11.9% in pCR vs 68.9% ± 15.4% in non-pCR, P=0.03). However, in patients with repeat PET after completion of PST, there were no statistical differences of these values (post-treatment SUVp, 1.09 ± 0.63 in pCR vs 1.29 ± 0.36 in non-pCR, P=0.42; RR of the SUVp, 83.7% ± 14.0% in pCR vs 67.5% ± 21.1% in non-pCR, P=0.17)

Conclusions: This study demonstrated that repeat PET after the 2nd cycle of PST, not after completion of PST could predict pCR in stage II/III breast cancer with preoperative chemotherapy. Acknowledgement NCC Grant #0910320.

Citation Information: Cancer Res 2010;70(24 Suppl):Abstract nr P2-09-01.

Characterization of the distal tail fiber locus and determination of the receptor for phage AR1, which specifically infects Escherichia coli O157:H7

Phage AR1 is similar to phage T4 in several essential genes but differs in host range. AR1 infects various isolates of Escherichia coli O157:H7 but does not infect K-12 strains that are commonly infected by T4. We report here the determinants that confer this infection specificity. In T-even phages, gp37 and gp38 are components of the tail fiber that are critical for phage-host interaction. The counterparts in AR1 may be similarly important and, therefore, were characterized. The AR1 gp37 has a sequence that differs totally from those of T2 and T4, except for a short stretch at the N terminus. The gp38 sequence, however, has some conservation between AR1 and T2 but not between AR1 and T4. The sequences that are most closely related to the AR1 gp37 and gp38 are those of phage Ac3 in the T2 family. To identify the AR1-specific receptor, E. coli O157:H7 was mutated by Tn10 insertion and selected for an AR1-resistant phenotype. A mutant so obtained has an insertion occurring at ompC that encodes an outer membrane porin. To confirm the role of OmpC in the AR1 infection, homologous replacement was used to create an ompC disruption mutant (RM). When RM was complemented with OmpC originated from an O157:H7 strain, but not from K-12, its AR1 susceptibility was fully restored. Our results suggest that the host specificity of AR1 is mediated at least in part through the OmpC molecule.

Characterization of a strain-specific monoclonal antibody to hepatitis delta virus antigen

Sequences of the hepatitis delta virus (HDV) vary to different degrees among isolates. A monoclonal antibody, designated as HP6A1, against the antigen of HDV (HDAg) has been characterized for its specificity. HP6A1 bound to HDAg of isolate 25 (genotype I) that was used for immunization, but not to others of both genotypes I and II. The epitope recognized by HP6A1 was then determined by a phage library displaying various heptapeptides. A consensus peptide deduced has the best match with that of residues 4-10 of HDAg (isolate 25). To confirm the phage mapping result, Escherichia coli recombinant proteins containing different lengths and various segments of HDAg (isolate 25) were constructed. The shortest HDAg segment contained in the fusion protein that reacted with HP6A1 was residues 1-10. When this peptide was added to the N-terminus of a heterologous protein engineered for eucaryotic expression, the fusion protein was detected by HP6A1. It is concluded that HP6A1 recognizes an epitope located at the N-terminus of HDAg (isolate 25). Since viruses of quasi-species exist in natural infections, a question of how different viral strains interact in vivo remains to be explored. The highly specific MAb opens a possibility to examine the fate of one strain in the presence of other related species in a cell transfection system.