Automated Detection of Brain Metastases on T1-Weighted MRI Using a Convolutional Neural Network: Impact of Volume Aware Loss and Sampling Strategy.

BACKGROUND: Detection of brain metastases (BM) and segmentation for treatment planning could be optimized with machine learning methods. Convolutional neural networks (CNNs) are promising, but their trade-offs between sensitivity and precision frequently lead to missing small lesions. HYPOTHESIS: Combining volume aware (VA) loss function and sampling strategy could improve BM detection sensitivity. STUDY TYPE: Retrospective. POPULATION: A total of 530 radiation oncology patients (55% women) were split into a training/validation set (433 patients/1460 BM) and an independent test set (97 patients/296 BM). FIELD STRENGTH/SEQUENCE: 1.5 T and 3 T, contrast-enhanced three-dimensional (3D) T1-weighted fast gradient echo sequences. ASSESSMENT: Ground truth masks were based on radiotherapy treatment planning contours reviewed by experts. A U-Net inspired model was trained. Three loss functions (Dice, Dice + boundary, and VA) and two sampling methods (label and VA) were compared. Results were reported with Dice scores, volumetric error, lesion detection sensitivity, and precision. A detected voxel within the ground truth constituted a true positive. STATISTICAL TESTS: McNemar's exact test to compare detected lesions between models. Pearson's correlation coefficient and Bland-Altman analysis to compare volume agreement between predicted and ground truth volumes. Statistical significance was set at P ≤ 0.05. RESULTS: Combining VA loss and VA sampling performed best with an overall sensitivity of 91% and precision of 81%. For BM in the 2.5-6 mm estimated sphere diameter range, VA loss reduced false negatives by 58% and VA sampling reduced it further by 30%. In the same range, the boundary loss achieved the highest precision at 81%, but a low sensitivity (24%) and a 31% Dice loss. DATA CONCLUSION: Considering BM size in the loss and sampling function of CNN may increase the detection sensitivity regarding small BM. Our pipeline relying on a single contrast-enhanced T1-weighted MRI sequence could reach a detection sensitivity of 91%, with an average of only 0.66 false positives per scan. EVIDENCE LEVEL: 3 TECHNICAL EFFICACY: Stage 2.

The impact of different stereotactic radiation therapy regimens for brain metastases on local control and toxicity.

PURPOSE: Stereotactic radiation therapy (SRT) enables focused, short course, high dose per fraction radiation delivery to brain tumors that are less ideal for single fraction treatment because of size, shape, or close proximity to sensitive structures. We sought to identify optimal SRT treatment regimens for maximizing local control while minimizing morbidity. METHODS AND MATERIALS: We performed a retrospective review of patients treated with SRT for solid brain metastases using variable dose schedules between 2001 and 2011 at 3 academic hospitals. Endpoints included (1) local control, (2) acute toxicity (Common Toxicity Criteria for Adverse Events v3.0), and (3) symptomatic radionecrosis. Kaplan-Meier and a competing risks methodology were used to estimate the actuarial rate of local failure and assess the association of clinical and treatment covariates with time to local failure. RESULTS: A total of 156 patients was identified. Common tumor histologies included breast (21%), non-small cell lung (32%), melanoma (22%), small cell lung (9%), and renal cell carcinoma (6%). The majority of lesions were supratentorial (57%). Median target volume was 3.99 mL (range, 0.04-58.42). Median total SRT dose was 25 Gy (range, 12-36), median fractional dose was 5 Gy (range, 2.5-11), and median number of fractions was 5 (range, 2-10). Cumulative incidence of local progression at 3, 6, 12, 18, and 24 months was 11%, 22%, 29%, 34%, and 36%. Total prescription dose was the only factor significantly associated with time to local progression on univariate (P = .02) and multivariable analysis (P = .01, adjusted hazards ratio, 0.87). Five patients experienced seizures within 10 days of SRT and 5 patients developed radionecrosis. All patients with documented radionecrosis received prior radiation to the index lesion. CONCLUSIONS: Our series of SRT for brain metastases found total prescription dose to be the only factor associated with local control. Both acute and long-term toxicity events from SRT were modest.

Local control after fractionated stereotactic radiation therapy for brain metastases.

Stereotactic radiosurgery (SRS) is frequently used in the management of brain metastases, but concerns over potential toxicity limit applications for larger lesions or those in eloquent areas. Fractionated stereotactic radiation therapy (SRT) is often substituted for SRS in these cases. We retrospectively analyzed the efficacy and toxicity outcomes of patients who received SRT at our institution. Seventy patients with brain metastases treated with SRT from 2006-2012 were analyzed. The rates of local and distant intracranial progression, overall survival, acute toxicity, and radionecrosis were determined. The SRT regimen was 25 Gy in 5 fractions among 87 % of patients. The most common tumor histologies were non-small cell lung cancer (37 %), breast cancer (20 %) and melanoma (20 %), and the median tumor diameter was 1.7 cm (range 0.4-6.4 cm). Median survival after SRT was 10.7 months. Median time to local progression was 17 months, with a local control rate of 68 % at 6 months and 56 % at 1 year. Acute toxicity was seen in 11 patients (16 %), mostly grade 1 or 2 with the most common symptom being mild headache. Symptomatic radiation-induced treatment change was seen on follow-up MRIs in three patients (4.3 %). SRT appears to be a safe and reasonably effective technique to treat brain metastases deemed less suitable for SRS, though dose intensification strategies may further improve local control.