Here we develop a tool to predict resectability of HER2+ breast cancer at breast conservation surgery (BCS) utilizing features identified on preoperative breast MRI. We identified patients with HER2+ breast cancer who obtained pre-operative breast MRI and underwent BCS between 2002-2013. From the contoured tumor on pre-operative MRI, shape, histogram, and co-occurrence and size zone matrix texture features were extracted. In univariate analysis, Spearmans correlation coefficient (Rs) was used to assess the correlation between each image feature and an endpoint (surgical re-excision). For multivariate modeling, we employed a support vector machine (SVM) method in a manner of leave-one-out cross-validation (LOOCV). Of 109 patients with HER2+breast cancer who underwent BCS, 39% underwent surgical re-excision. 62% had residual cancer at re-excision. In univariate analysis, solidity (Rs = -0.32, p = 0.009) and extent (Rs = -0.29, p = 0.019) were significantly associated with re-excision. Skewness in post-contrast 1, 2, and 3 (Rs = 0.25, p = 0.045; Rs = 0.30, p = 0.015; Rs = 0.28, p = 0.026) and kurtosis in post-contrast 1 (Rs = 0.26, p = 0.035) were also statistically significant. LOOCV-based SVM test achieved 74.4% specificity and 71.4% sensitivity when 21 features were used. Thus, tumor texture, histogram and morphological MRI features may assist surgical planning, encouraging wide margins or mastectomy in patients who may otherwise go on to re-excision.

PURPOSE: To use features extracted from magnetic resonance (MR) images and a machine-learning method to assist in differentiating breast cancer molecular subtypes. MATERIALS AND METHODS: This retrospective Health Insurance Portability and Accountability Act (HIPAA)-compliant study received Institutional Review Board (IRB) approval. We identified 178 breast cancer patients between 2006-2011 with: 1) ERPR + (n = 95, 53.4%), ERPR-/HER2 + (n = 35, 19.6%), or triple negative (TN, n = 48, 27.0%) invasive ductal carcinoma (IDC), and 2) preoperative breast MRI at 1.5T or 3.0T. Shape, texture, and histogram-based features were extracted from each tumor contoured on pre- and three postcontrast MR images using in-house software. Clinical and pathologic features were also collected. Machine-learning-based (support vector machines) models were used to identify significant imaging features and to build models that predict IDC subtype. Leave-one-out cross-validation (LOOCV) was used to avoid model overfitting. Statistical significance was determined using the Kruskal-Wallis test. RESULTS: Each support vector machine fit in the LOOCV process generated a model with varying features. Eleven out of the top 20 ranked features were significantly different between IDC subtypes with P < 0.05. When the top nine pathologic and imaging features were incorporated, the predictive model distinguished IDC subtypes with an overall accuracy on LOOCV of 83.4%. The combined pathologic and imaging models accuracy for each subtype was 89.2% (ERPR+), 63.6% (ERPR-/HER2+), and 82.5% (TN). When only the top nine imaging features were incorporated, the predictive model distinguished IDC subtypes with an overall accuracy on LOOCV of 71.2%. The combined pathologic and imaging models accuracy for each subtype was 69.9% (ERPR+), 62.9% (ERPR-/HER2+), and 81.0% (TN). CONCLUSION: We developed a machine-learning-based predictive model using features extracted from MRI that can distinguish IDC subtypes with significant predictive power. J. Magn. Reson. Imaging 2016;44:122-129.

PURPOSE: To investigate the association between a validated, gene-expression-based, aggressiveness assay, Oncotype Dx RS, and morphological and texture-based image features extracted from magnetic resonance imaging (MRI). MATERIALS AND METHODS: This retrospective study received Internal Review Board approval and need for informed consent was waived. Between 2006-2012, we identified breast cancer patients with: 1) ER+, PR+, and HER2- invasive ductal carcinoma (IDC); 2) preoperative breast MRI; and 3) Oncotype Dx RS test results. Extracted features included morphological, histogram, and gray-scale correlation matrix (GLCM)-based texture features computed from tumors contoured on pre- and three postcontrast MR images. Linear regression analysis was performed to investigate the association between Oncotype Dx RS and different clinical, pathologic, and imaging features. P < 0.05 was considered statistically significant. RESULTS: Ninety-five patients with IDC were included with a median Oncotype Dx RS of 16 (range: 0-45). Using stepwise multiple linear regression modeling, two MR-derived image features, kurtosis in the first and third postcontrast images and histologic nuclear grade, were found to be significantly correlated with the Oncotype Dx RS with P = 0.0056, 0.0005, and 0.0105, respectively. The overall model resulted in statistically significant correlation with Oncotype Dx RS with an R-squared value of 0.23 (adjusted R-squared = 0.20; P = 0.0002) and a Spearmans rank correlation coefficient of 0.49 (P < 0.0001). CONCLUSION: A model for IDC using imaging and pathology information correlates with Oncotype Dx RS scores, suggesting that image-based features could also predict the likelihood of recurrence and magnitude of chemotherapy benefit.

Background

Cerebral autosomal-dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) is the most common genetic cause of stroke. In addition to ischemic stroke, CADASIL predisposes to development of cerebral microbleeds (CMB). CMB and hypertension are known to be associated with intracerebral hemorrhage (ICH). The purpose of this study was to analyze the relationships among CMB, hypertension, and ICH in CADASIL.

Materials and methods

We enrolled 94 genetically confirmed CADASIL patients from 76 unrelated families at Jeju National University Hospital (Korea) between March 2012 and February 2015. We analyzed CMB presence, number, and distribution on susceptibility-weighted imaging MRI using the microbleed anatomical rating scale. Multiple logistic regression was used to determine factors associated with the presence of CMB and ICH.

Results

CMB were observed in 62 patients (66%), median number of CMB per patient was 4 (range 0-121). Twenty-two ICHs were found in 16 patients (17%). There was incongruence between the most common site of CMB (thalamus) and that of ICH (basal ganglia). Hypertension was independently associated with the presence of CMB (multiple regression OR, 2.71; 95% CI 1.02-7.18, p < 0.05), and CMB ≥ 9 (highest third) was significantly associated with the presence of ICH (multiple regression OR = 9.50, 95% CI 1.08-83.71, p < 0.05).

Conclusion

In this CADASIL sample, presence of hypertension was independently associated with CMB presence, and CMB burden was independently associated with ICH. Incongruence of sites for CMB and ICH is currently unexplained and requires further study.

Background

Cerebral autosomal-dominant arteriopathy with subcortical infarcts and leukoencephalopathy (CADASIL) is the most common genetic cause of stroke. In addition to ischemic stroke, CADASIL predisposes to development of cerebral microbleeds (CMB). CMB and hypertension are known to be associated with intracerebral hemorrhage (ICH). The purpose of this study was to analyze the relationships among CMB, hypertension, and ICH in CADASIL.

Materials and methods

We enrolled 94 genetically confirmed CADASIL patients from 76 unrelated families at Jeju National University Hospital (Korea) between March 2012 and February 2015. We analyzed CMB presence, number, and distribution on susceptibility-weighted imaging MRI using the microbleed anatomical rating scale. Multiple logistic regression was used to determine factors associated with the presence of CMB and ICH.

Results

CMB were observed in 62 patients (66%), median number of CMB per patient was 4 (range 0-121). Twenty-two ICHs were found in 16 patients (17%). There was incongruence between the most common site of CMB (thalamus) and that of ICH (basal ganglia). Hypertension was independently associated with the presence of CMB (multiple regression OR, 2.71; 95% CI 1.02-7.18, p < 0.05), and CMB ≥ 9 (highest third) was significantly associated with the presence of ICH (multiple regression OR = 9.50, 95% CI 1.08-83.71, p < 0.05).

Conclusion

In this CADASIL sample, presence of hypertension was independently associated with CMB presence, and CMB burden was independently associated with ICH. Incongruence of sites for CMB and ICH is currently unexplained and requires further study.

Purpose: To demonstrate that a mathematical model can be used to quantitatively understand tumor cellular dynamics during a course of radiotherapy and to predict the likelihood of local control as a function of dose and treatment fractions.Experimental Design: We model outcomes for early-stage, localized non-small cell lung cancer (NSCLC), by fitting a mechanistic, cellular dynamics-based tumor control probability that assumes a constant local supply of oxygen and glucose. In addition to standard radiobiological effects such as repair of sub-lethal damage and the impact of hypoxia, we also accounted for proliferation as well as radiosensitivity variability within the cell cycle. We applied the model to 36 published and two unpublished early-stage patient cohorts, totaling 2,701 patients.Results: Precise likelihood best-fit values were derived for the radiobiological parameters: α [0.305 Gy^-1; 95% confidence interval (CI), 0.120-0.365], the α/β ratio (2.80 Gy; 95% CI, 0.40-4.40), and the oxygen enhancement ratio (OER) value for intermediately hypoxic cells receiving glucose but not oxygen (1.70; 95% CI, 1.55-2.25). All fractionation groups are well fitted by a single dose-response curve with a high χ² P value, indicating consistency with the fitted model. The analysis was further validated with an additional 23 patient cohorts (n = 1,628). The model indicates that hypofractionation regimens overcome hypoxia (and cell-cycle radiosensitivity variations) by the sheer impact of high doses per fraction, whereas lower dose-per-fraction regimens allow for reoxygenation and corresponding sensitization, but lose effectiveness for prolonged treatments due to proliferation.Conclusions: This proposed mechanistic tumor-response model can accurately predict overtreatment or undertreatment for various treatment regimens. Clin Cancer Res; 23(18); 5469-79. ©2017 AACR.

Background and purpose

To study internal and external generalizability of temporal dose-response relationships for xerostomia after intensity-modulated radiotherapy (IMRT) for head and neck cancer, and to investigate potential amendments of the QUANTEC guidelines.

Material and methods

Objective xerostomia was assessed in 121 patients (n_Cohort1=55; n_Cohort2=66) treated to 70Gy@2Gy in 2006-2015. Univariate and multivariate analyses (UVA, MVA with 1000 bootstrap populations) were conducted in Cohort1, and generalizability of the best-performing MVA model was investigated in Cohort2 (performance: AUC, p-values, and Hosmer-Lemeshow p-values (p_HL)). Ultimately and for clinical guidance, minimum mean dose thresholds to the contralateral and the ipsilateral parotid glands (Dmean_contra, Dmean_ipsi) were estimated from the generated dose-response curves.

Results

The observed xerostomia rate was 38%/47% (3months) and 19%/23% (11-12months) in Cohort1/Cohort2. Risk of xerostomia at 3months increased for higher Dmean_contra and Dmean_ipsi (Cohort1: 0.17·Dmean_contra+0.11·Dmean_ipsi-8.13; AUC=0.90±0.05; p=0.0002±0.002; p_HL=0.22±0.23; Cohort2: AUC=0.81; p<0.0001; p_HL=0.27). The identified minimum Dmean_contra thresholds were lower than in the QUANTEC guidelines (Cohort1/Cohort2: Dmean_contra=12/19Gy; Dmean_contra, Dmean_ipsi=16, 25/20, 26Gy).

Conclusions

Increased Dmean_contra and Dmean_ipsi explain short-term xerostomia following IMRT. Our results also suggest decreasing Dmean_contra to below 20Gy, while keeping Dmean_ipsi to around 25Gy. Long-term xerostomia was less frequent, and no dose-response relationship was established for this follow-up time.

Hirschsprung disease (HSCR) is a congenital and heterogeneous disorder characterized by the absence of intramural nervous plexuses along variable lengths of the hindgut. Although RET is a well-established risk factor, a recent genome-wide association study (GWAS) of HSCR has identified NRG1 as an additional susceptibility locus. To discover additional risk loci, we performed a GWAS of 123 sporadic HSCR patients and 432 unaffected controls using a large-scale platform with coverage of over 1 million polymorphic markers. The result was that our study replicated the findings of RET-CSGALNACT2-RASGEF1A genomic region (rawP = 5.69×10(-19) before a Bonferroni correction; corrP = 4.31×10(-13) after a Bonferroni correction) and NRG1 as susceptibility loci. In addition, this study identified SLC6A20 (adjP = 2.71×10(-6)), RORA (adjP = 1.26×10(-5)), and ABCC9 (adjP = 1.86×10(-5)) as new potential susceptibility loci under adjusting the already known loci on the RET-CSGALNACT2-RASGEF1A and NRG1 regions, although none of the SNPs in these genes passed the Bonferroni correction. In further subgroup analysis, the RET-CSGALNACT2-RASGEF1A genomic region was observed to have different significance levels among subgroups: short-segment (S-HSCR, corrP = 1.71×10(-5)), long-segment (L-HSCR, corrP = 6.66×10(-4)), and total colonic aganglionosis (TCA, corrP>0.05). This differential pattern in the significance level suggests that other genomic loci or mechanisms may affect the length of aganglionosis in HSCR subgroups during enteric nervous system (ENS) development. Although functional evaluations are needed, our findings might facilitate improved understanding of the mechanisms of HSCR pathogenesis.

Background

The aims of this study were to investigate temporal patterns and potential risk factors for severe hyposalivation (xerostomia) after intensity-modulated radiotherapy (IMRT) for head and neck cancer (HNC), and to test the two QUANTEC (Quantitative Analysis of Normal Tissue Effects in the Clinic) guidelines.

Patients and methods

Sixty-three patients treated at the Memorial Sloan Kettering Cancer Center between 2006 and 2015, who had a minimum of three stimulated whole mouth saliva flow measurements (WMSFM) at a median follow-up time of 11 (range: 3-24) months were included. Xerostomia was defined as WMSFM ≤25% compared to relative pre-radiotherapy. Patients were stratified into three follow-up groups: 1: <6 months; 2: 6-11 months; and 3: 12-24 months. Potential risk factors were investigated (Mann-Whitney U test), and relative risks (RRs) assessed for the two QUANTEC guidelines.

Results

The incidence of xerostomia was 27%, 14% and 17% at follow-up time points 1, 2 and 3, respectively. At <6 months, the mean dose to the contralateral and the ipsilateral parotid glands (Dmean_contra, Dmean_ipsi) was higher among patients with xerostomia (Dmean_contra: 25 Gy vs. 15 Gy; Dmean_ipsi: 44 Gy vs. 25 Gy). Patients with xerostomia had higher pre-RT WMSFM (3.5 g vs. 2.4 g), and had been treated more frequently with additional chemotherapy (93% vs. 63%; all 4 variables: p < 0.05). At 6-11 months, Dmean_contra among patients with xerostomia was higher compared to patients without (26 Gy vs. 20 Gy). The RR as specified by the one- and two-gland QUANTEC guideline was 2.3 and 1.4 for patients with <6 months follow-up time, and 2.0 and 1.2 for patients with longer follow-up (6-11 + 6-24 months).

Conclusion

Xerostomia following IMRT peaks within six months post-radiotherapy and fades with time. Limiting the mean dose to both parotid glands (ipsilateral <25 Gy, contralateral <25 Gy) and reducing the use of chemotherapy will likely decrease the rate of xerostomia. Both QUANTEC guidelines are effective in preventing xerostomia.

With the era of big data, the utilization of machine learning algorithms in radiation oncology is rapidly growing with applications including: treatment response modeling, treatment planning, contouring, organ segmentation, image-guidance, motion tracking, quality assurance, and more. Despite this interest, practical clinical implementation of machine learning as part of the day-to-day clinical operations is still lagging. The aim of this white paper is to further promote progress in this new field of machine learning in radiation oncology by highlighting its untapped advantages and potentials for clinical advancement, while also presenting current challenges and open questions for future research. The targeted audience of this paper includes newcomers as well as practitioners in the field of medical physics/radiation oncology. The paper also provides general recommendations to avoid common pitfalls when applying these powerful data analytic tools to medical physics and radiation oncology problems and suggests some guidelines for transparent and informative reporting of machine learning results.