Deploying deep learning models on unseen medical imaging using adversarial domain adaptation.

The fundamental challenge in machine learning is ensuring that trained models generalize well to unseen data. We developed a general technique for ameliorating the effect of dataset shift using generative adversarial networks (GANs) on a dataset of 149,298 handwritten digits and dataset of 868,549 chest radiographs obtained from four academic medical centers. Efficacy was assessed by comparing area under the curve (AUC) pre- and post-adaptation. On the digit recognition task, the baseline CNN achieved an average internal test AUC of 99.87% (95% CI, 99.87-99.87%), which decreased to an average external test AUC of 91.85% (95% CI, 91.82-91.88%), with an average salvage of 35% from baseline upon adaptation. On the lung pathology classification task, the baseline CNN achieved an average internal test AUC of 78.07% (95% CI, 77.97-78.17%) and an average external test AUC of 71.43% (95% CI, 71.32-71.60%), with a salvage of 25% from baseline upon adaptation. Adversarial domain adaptation leads to improved model performance on radiographic data derived from multiple out-of-sample healthcare populations. This work can be applied to other medical imaging domains to help shape the deployment toolkit of machine learning in medicine.

Population scale latent space cohort matching for the improved use and exploration of observational trial data.

A significant amount of clinical research is observational by nature and derived from medical records, clinical trials, and large-scale registries. While there is no substitute for randomized, controlled experimentation, such experiments or trials are often costly, time consuming, and even ethically or practically impossible to execute. Combining classical regression and structural equation modeling with matching techniques can leverage the value of observational data. Nevertheless, identifying variables of greatest interest in high-dimensional data is frequently challenging, even with application of classical dimensionality reduction and/or propensity scoring techniques. Here, we demonstrate that projecting high-dimensional medical data onto a lower-dimensional manifold using deep autoencoders and post-hoc generation of treatment/control cohorts based on proximity in the lower-dimensional space results in better matching of confounding variables compared to classical propensity score matching (PSM) in the original high-dimensional space (P<0.0001) and performs similarly to PSM models constructed by experts with prior knowledge of the underlying pathology when evaluated on predicting risk ratios from real-world clinical data. Thus, in cases when the underlying problem is poorly understood and the data is high-dimensional in nature, matching in the autoencoder latent space might be of particular benefit.

Time on Therapy for at Least Three Months Correlates with Overall Survival in Metastatic Renal Cell Carcinoma.

With 15 drugs currently approved for the treatment of metastatic renal cell carcinoma (mRCC) and even more combination regimens with immunotherapy on the horizon, there remains a distinct lack of molecular biomarkers for therapeutic efficacy. Our study reports on real-world clinical outcomes of mRCC patients from a tertiary academic medical center treated with empirically selected standard-of-care therapy. We utilized the Stanford Renal Cell Carcinoma Database (RCCD) to report on various outcome measures, including overall survival (OS) and the median number of lines of targeted therapies received from the time of metastatic diagnosis. We found that most metastatic patients did not survive long enough to attempt even half of the available targeted therapies. We also noted that patients who failed to receive a clinical benefit within the first two lines of therapy could still go on to experience clinical benefit in later lines of therapy. The term, "clinical benefit" was assigned to a line of therapy if a patient remained on drug treatment for three months or longer. Moreover, patients with clinical benefit in at least one line of therapy experienced significantly longer OS compared to those who did not have clinical benefit in at least one line of therapy. Developing biomarkers that identify patients who will receive clinical benefit in individual lines of therapy is one potential strategy for achieving rational drug sequencing in mRCC.

The 'Achilles Heel' of Metabolism in Renal Cell Carcinoma: Glutaminase Inhibition as a Rational Treatment Strategy.

An important hallmark of cancer is 'metabolic reprogramming' or the rewiring of cellular metabolism to support rapid cell proliferation [1-5]. Metabolic reprogramming through oncometabolite-mediated transformation or activation of oncogenes in renal cell carcinoma (RCC) globally impacts energy production as well as glucose and glutamine utilization in RCC cells, which can promote dependence on glutamine supply to support cell growth and proliferation [6, 7]. Novel inhibitors of glutaminase, a key enzyme in glutamine metabolism, target glutamine addiction as a viable treatment strategy in metastatic RCC (mRCC). Here, we review glutamine metabolic pathways and how changes in cellular glutamine utilization enable the progression of RCC. This overview provides scientific rationale for targeting this pathway in patients with mRCC. We will summarize the current understanding of cellular and molecular mechanisms underlying anti-tumor efficacy of glutaminase inhibitors in RCC, provide an overview of clinical efforts targeting glutaminase in mRCC, and review approaches for identifying biomarkers for patient stratification and detecting therapeutic response early on in patients treated with this novel class of anti-cancer drug. Ultimately, results of ongoing clinical trials will demonstrate whether glutaminase inhibition can be a worthy addition to the current armamentarium of drugs used for patients with mRCC.

Five fraction image-guided radiosurgery for primary and recurrent meningiomas.

PURPOSE: Benign tumors that arise from the meninges can be difficult to treat due to their potentially large size and proximity to critical structures such as cranial nerves and sinuses. Single fraction radiosurgery may increase the risk of symptomatic peritumoral edema. In this study, we report our results on the efficacy and safety of five fraction image-guided radiosurgery for benign meningiomas. MATERIALS/METHODS: Clinical and radiographic data from 38 patients treated with five fraction radiosurgery were reviewed retrospectively. Mean tumor volume was 3.83 mm(3) (range, 1.08-20.79 mm(3)). Radiation was delivered using the CyberKnife, a frameless robotic image-guided radiosurgery system with a median total dose of 25 Gy (range, 25-35 Gy). RESULTS: The median follow-up was 20 months. Acute toxicity was minimal with eight patients (21%) requiring a short course of steroids for headache at the end of treatment. Pre-treatment neurological symptoms were present in 24 patients (63.2%). Post treatment, neurological symptoms resolved completely in 14 patients (58.3%), and were persistent in eight patients (33.3%). There were no local failures, 24 tumors remained stable (64%) and 14 regressed (36%). Pre-treatment peritumoral edema was observed in five patients (13.2%). Post-treatment asymptomatic peritumoral edema developed in five additional patients (13.2%). On multivariate analysis, pre-treatment peritumoral edema and location adjacent to a large vein were significant risk factors for radiographic post-treatment edema (p = 0.001 and p = 0.026 respectively). CONCLUSION: These results suggest that five fraction image-guided radiosurgery is well tolerated with a response rate for neurologic symptoms that is similar to other standard treatment options. Rates of peritumoral edema and new cranial nerve deficits following five fraction radiosurgery were low. Longer follow-up is required to validate the safety and long-term effectiveness of this treatment approach.

Hypofractionated stereotactic body radiation therapy as monotherapy for intermediate-risk prostate cancer.

BACKGROUND: Hypofractionated stereotactic body radiation therapy (SBRT) has been advanced as monotherapy for low-risk prostate cancer. We examined the dose distributions and early clinical outcomes using this modality for the treatment of intermediate-risk prostate cancer. METHODS: Forty-one sequential hormone-naïve intermediate-risk prostate cancer patients received 35-36.25 Gy of CyberKnife-delivered SBRT in 5 fractions. Radiation dose distributions were analyzed for coverage of potential microscopic ECE by measuring the distance from the prostatic capsule to the 33 Gy isodose line. PSA levels, toxicities, and quality of life (QOL) measures were assessed at baseline and follow-up. RESULTS: All patients completed treatment with a mean coverage by the 33 Gy isodose line extending >5 mm beyond the prostatic capsule in all directions except posteriorly. Clinical responses were documented by a mean PSA decrease from 7.67 ng/mL pretreatment to 0.64 ng/mL at the median follow-up of 21 months. Forty patients remain free from biochemical progression. No Grade 3 or 4 toxicities were observed. Mean EPIC urinary irritation/obstruction and bowel QOL scores exhibited a transient decline post-treatment with a subsequent return to baseline. No significant change in sexual QOL was observed. CONCLUSIONS: In this intermediate-risk patient population, an adequate radiation dose was delivered to areas of expected microscopic ECE in the majority of patients. Although prospective studies are needed to confirm long-term tumor control and toxicity, the short-term PSA response, biochemical relapse-free survival rate, and QOL in this interim analysis are comparable to results reported for prostate brachytherapy or external beam radiotherapy. TRIAL REGISTRATION: The Georgetown Institutional Review Board has approved this retrospective study (IRB 2009-510).

CyberKnife with Tumor Tracking: An Effective Treatment for High-Risk Surgical Patients with Stage I Non-Small Cell Lung Cancer.

Published data suggests that wedge resection for stage I non-small cell lung cancer (NSCLC) is associated with improved overall survival compared to stereotactic body radiation therapy. We report CyberKnife outcomes for high-risk surgical patients with biopsy-proven stage I NSCLC. PET/CT imaging was completed for staging. Three-to-five gold fiducial markers were implanted in or near tumors to serve as targeting references. Gross tumor volumes (GTVs) were contoured using lung windows; the margins were expanded by 5 mm to establish the planning treatment volume (PTV). Treatment plans were designed using a mean of 156 pencil beams. Doses delivered to the PTV ranged from 42 to 60 Gy in three fractions. The 30 Gy isodose contour extended at least 1 cm from the GTV to eradicate microscopic disease. Treatments were delivered using the CyberKnife system with tumor tracking. Examination and PET/CT imaging occurred at 3 month follow-up intervals. Forty patients (median age 76) with a median maximum tumor diameter of 2.6 cm (range, 1.4-5.0 cm) and a mean post-bronchodilator percent predicted forced expiratory volume in 1 s (FEV1) of 57% (range, 21-111%) were treated. A median dose of 48 Gy was delivered to the PTV over 3-13 days (median, 7 days). The 30 Gy isodose contour extended a mean 1.9 cm from the GTV. At a median 44 months (range, 12-72 months) follow-up, the 3 year Kaplan-Meier locoregional control and overall survival estimates compare favorably with contemporary wedge resection outcomes at 91 and 75%, respectively. CyberKnife is an effective treatment approach for stage I NSCLC that is similar to wedge resection, eradicating tumors with 1-2 cm margins in order to preserve lung function. Prospective randomized trials comparing CyberKnife with wedge resection are necessary to confirm equivalence.

Low incidence of new biochemical and clinical hypogonadism following hypofractionated stereotactic body radiation therapy (SBRT) monotherapy for low- to intermediate-risk prostate cancer.

BACKGROUND: The CyberKnife is an appealing delivery system for hypofractionated stereotactic body radiation therapy (SBRT) because of its ability to deliver highly conformal radiation therapy to moving targets. This conformity is achieved via 100s of non-coplanar radiation beams, which could potentially increase transitory testicular irradiation and result in post-therapy hypogonadism. We report on our early experience with CyberKnife SBRT for low- to intermediate-risk prostate cancer patients and assess the rate of inducing biochemical and clinical hypogonadism. METHODS: Twenty-six patients were treated with hypofractionated SBRT to a dose of 36.25 Gy in 5 fractions. All patients had histologically confirmed low- to intermediate-risk prostate adenocarcinoma (clinical stage ≤ T2b, Gleason score ≤ 7, PSA ≤ 20 ng/ml). PSA and total testosterone levels were obtained pre-treatment, 1 month post-treatment and every 3 months thereafter, for 1 year. Biochemical hypogonadism was defined as a total serum testosterone level below 8 nmol/L. Urinary and gastrointestinal toxicity was assessed using Common Toxicity Criteria v3; quality of life was assessed using the American Urological Association Symptom Score, Sexual Health Inventory for Men and Expanded Prostate Cancer Index Composite questionnaires. RESULTS: All 26 patients completed the treatment with a median 15 months (range, 13-19 months) follow-up. Median pre-treatment PSA was 5.75 ng/ml (range, 2.3-10.3 ng/ml), and a decrease to a median of 0.7 ng/ml (range, 0.2-1.8 ng/ml) was observed by one year post-treatment. The median pre-treatment total serum testosterone level was 13.81 nmol/L (range, 5.55 - 39.87 nmol/L). Post-treatment testosterone levels slowly decreased with the median value at one year follow-up of 10.53 nmol/L, significantly lower than the pre-treatment value (p < 0.013). The median absolute fall was 3.28 nmol/L and the median percent fall was 23.75%. There was no increase in biochemical hypogonadism at one year post-treatment. Average EPIC sexual and hormonal scores were not significantly changed by one year post-treatment. CONCLUSIONS: Hypofractionated SBRT offers the radiobiological benefit of a large fraction size and is well-tolerated by men with low- to intermediate-risk prostate cancer. Early results are encouraging with an excellent biochemical response. The rate of new biochemical and clinical hypogonadism was low one year after treatment.