How to Reduce Missed Diagnosis of Breast Cancer Heeboong Park M.D., Ph.D. Park Surgical Clinic Medical Park Co., Ltd.
Definition of Missed Diagnosis Misdiagnosis means that a doctor has diagnosed a person with the wrong medical condition. Missed Diagnosis: Failure to diagnose or delayed diagnosis, occurs when a doctor fails to identify a medical condition at the time they re presented with it. Objective, whole breast examination - mammography After multimodal diagnosis of breast cancer - diagnostic sensitivity of examination modality Breast cancer detected after recent examination is suspected to missed diagnosis.
Missed Diagnosis PE Mammography US MRI Biopsy
Interval breast cancer The definition of interval cancers in the NHSBSP (NHS Breast Screening Radiologists Quality Assurance Committee, 2005) is consistent with that in the European guidelines (Perry et al, 2006); breast cancers diagnosed in the interval between scheduled screening episodes in women screened and given a normal' screening result that is, the previous screening episode was negative. In our analyses we defined core interval cancers as those occurring within 36 months of a woman's last negative-screening episode in women aged between 50 and 64 at their last routine screen; Bennett et al. Br J Cancer. 2011 Feb 15; 104(4): 571 577.
Breast cancer tumor growth estimated through mammography screening data Cancer incidence and tumor measurement data from 395,188 women aged 50 to 69 years. Tumor growth varied considerably between subjects, with 5% of tumors taking less than 1.2 months to grow from 10 mm to 20 mm in diameter, and another 5% taking more than 6.3 years. The mean time a tumor needed to grow from 10 mm to 20 mm in diameter was estimated as 1.7 years, increasing with age. The screen test sensitivity was estimated to increase sharply with tumor size, rising from 26% at 5 mm to 91% at 10 mm. Harald Weedon-Fekjæ et. al Breast Cancer Research 2008 10:R41
ESTIMATES OF BREAST CANCER GROWTH RATE FROM MAMMOGRAMS AND ITS RELATION TO TUMOUR CHARACTERISTICS. This study aimed to investigate the growth rate of 31 consecutive invasive breast cancers based on volume measures on at least two serial mammograms and its relation to histopathological findings. The average tumor volume-doubling time in all invasive breast cancer subtypes was 282 d (range 46-749 d). Grade III breast cancers had a significantly shorter average tumour volumedoubling time of 105 d (range 46-157 d) compared with Grade I and II tumors (average of 296 d, range 147-531 d and average of 353 d, range 139-749 d, respectively) (p = 0.002). Multiple linear regression identified that tumor volume-doubling time was positively associated with patient age, histological grade and progesterone receptor expression and inversely associated with axillary lymph node involvement, human epidermal growth factor receptor 2 and Ki-67 expression (p < 0.001). Förnvik D et al. Radiat Prot Dosimetry. 2016 Jun;169(1-4):151-7.
A Review of Interval Breast Cancers Diagnosed among Participants of the Nova Scotia Breast Screening Program The NSBSP maintains databases that contain information on women participating in the program. From 1991 to 2004, the NSBSP collected information for 302,234 screening examinations performed in 115,433 women. The rate of missed cancers per 1000 women screened was one-half of the true interval cancer rate among women screened annually (for ages 40 49 years, 0.45 vs 0.93; for ages 50 69 years, 1.08 vs 2.22). Among women aged 50 69 years who were screened biennially, the rate of missed cancers per 1000 women screened was one-third of the true interval cancer rate (0.90 vs 3.15). Similarly, the rate of missed cancers per 10 000 screening examinations was one-half of the true interval rate among those 40 49 years old (1.95 vs 3.99) and one-third of the true interval cancer rate among those 50 69 years old (3.34 vs 10.44). Jennifer I. Payne et al. Radiology. 2013 Jan;266(1):96-103.
Missed Breast Carcinoma; Why and How to Avoid? Study over a two-year interval 152 cases revealed 121 infiltrating ductal carcinomas, 2 lobular, 4 mucinous, 14 inflammatory carcinomas, 6 carcinomas in situ (3 of which were intracystic), 2 intraductal papillary carcinomas and 3 cases with Paget's disease of the nipple. 4 causative factors; patient, tumor, technical or provider factors Tumor factors were the most commonly encountered, accounting for 44.1%, while provider factors were the least commonly encountered in 14.5% Kamal et al. J Egyptian NCI, 19(3): 178-194, 2007
44 carcinomas were detected on double and re-reading by more experienced radiologists. Additional mammographic views were recommended in 35 (23%) cases. Complementary ultrasound examination was performed for all 152 cases (100%) and showed a higher sensitivity than mammography in carcinoma detection. It was diagnostic in 138 (90.8%) cases only. In the remaining 14 cases, further MRI and biopsy were performed.
Happy eye syndrome The observation of an obvious finding (benign or malignant) may cause the "happy eye syndrome," misleading the radiologist into not looking carefully for other lesions. Small cancer with big benign tumor- two cases Multicentric and multifocal carcinomas were missed in 4.6% of cases (7 patients) and contralateral carcinomas in 3 other patients (2% of cases).
How to avoid missing a breast carcinoma? Review clinical data and use US and other adjunct techniques as MRI and biopsy to assess a palpable or mammographically detected mass. Be strict about positioning and technical factors. Try to optimize image quality. Be alert to subtle features of breast cancers. Always consider the well defined carcinoma. Compare current images with multiple prior studies to look for subtle increases in lesion size. Look for other lesions when one abnormality is seen. Judge a lesion by its most malignant features. Double reading and the use of computer aided diagnosis (CAD) and finally FFDM (Full Field Digital Mammography). Close cooperation between the oncologist, radiologist and pathologist is essential to avoid missing any case of breast carcinoma.
Detection of breast cancer in asymptomatic and symptomatic groups using computer-aided detection with full-field digital mammography. Analyzed digital mammography and CAD images from 210 patients diagnosed with breast cancer. The detection rate of the CAD system was 87.8% in the asymptomatic group. The sensitivity in different tissue densities was 100% in P1, 88.9% in P2, 94.4% P3, and 66.7% in extremely dense breasts (P4). The detection rate of the CAD system in the symptomatic group was 87.2%, and the sensitivity was 90.5%, 90%, 86.6%, and 75% in P1-P4 breasts, respectively. In the asymptomatic group, the CAD system detected 90.3% of invasive ductal carcinomas, not otherwise specified (IDC-NOS) and 88.9% of ductal carcinomas in situ (DCIS), but did not detect other types of malignancy. In the symptomatic group, the CAD system detected 88.2% of IDC-NOS, 88.9% of DCIS and 75% of other types of malignancy. When analyzed according to tumor size, the sensitivity of CAD in the asymptomatic and symptomatic groups was 82.6% and 83.3% for tumors <1 cm, 76.5% and 82.4% for tumors between 1 and 2 cm, and 91.7% and 89% in tumors >2 cm. Park et al J Breast Cancer. 2013 Sep; 16(3): 322 328.
High-Resolution Breast Cancer Screening with Multi-View Deep Convolutional Neural Networks 129,208 patients, 201,698 exams and 886,437 images in the data set input image: 829k (57k) 2600 2000 target: -lesion: lesion classification (benign vs. malignant), -mass: mass detection -MC: microcalcification detection. Geraset et al arxiv:1703.07047v2 2017
Applying Data-driven Imaging Biomarker in Mammography for Breast Cancer Screening: Preliminary Study 29,107 digital mammograms from five institutions (4,339 cancer cases and 24,768 normal cases) were included Input size 1600x1600, SGD(lr=0.001, decay=0.0001, momentum=0.9) 30 epoch, minibatch=64(16 pt) 8 GPU Kim et al. Scientiic Reports 8:2762, 2018
Malignancy Detection on Mammography Using Dual Deep Convolutional Neural Networks and Genetically Discovered False Color Input Enhancement. Mammography sensitivity ranging from 62 to 87% and specificity from 75 to 91%. Inception v3, preprocessing and computer vision methods. The Digital Database for Screening Mammography (DDSM) 의 6,000 mammographic image and Zebra Mammography Dataset image 1,739. We also utilize dual deep convolutional neural networks at different scales, for classification of full mammogram images and derivative patches combined with a random forest gating network as a novel architectural solution capable of discerning malignancy with a sensitivity of 0.91 and a specificity of 0.80. Teare et al. J Digit Imaging. 2017 Aug;30(4):499-505
CAD Impact iview SonoEye TM Typical Benign Case Typical Malignant Case - ROI image vs. full size image - Biopsy-proven DB : about 1,000 cases - Malignant cases : about 400. - Benign cases : about 600 - In situations in which the radiologist and the computer disagree, this is useful solutions. - Computer-generated malignant score is calculated based on the retrieved images. - User can infer what's the basis of computer's malignancy score.
SonoEye 를이용한유방초음파검사상종괴에대한컴퓨터보조진단 (CAD) 의유용성 윤태일, 박신혜, 변종석, 박희붕 J. Breast Cancer 2006 9(2) 110pp 2004 May ~ 2005 Apr. Breast cancer 55 110 images Benign mass 79 156 images US-CAD : SonoEye, Cadimpact Sensitivity: 83.6 ~ 96.4% Specificity: 53.2 ~ 73.4 %
삼성메디슨유방과갑상선에서 AI 이용자동진단프로그램
A deep learning framework for supporting the classification of breast lesions in ultrasound images 5,151 patients cases containing a total of 7,408 ultrasound breast images. 4,254 benign and 3,154 malignant lesions. GoogleLeNet (22 layers, input 255x255, grey -> color) Data augmentation: 147 times(553,455 vs 413,658) SGD(lr=0.0001, decay=0.0002, momentum=0.9) 1 GPU, minibatch size =32 Accuracy of about 0.9 (90%), a sensitivity of 0.86 and a specificity of 0.96. Area under ROC curve AUC 0.9 Han et al. Phys Med Biol. 2017 Sep 15;62(19):7714-7728
Deep learning breast US(I) Malignant(invasive) vs benign neoplasm Training 5,095 validation 1,100 images from Park Surgical Clinic Keras with Tensorflow backend. Various models and accuracy -Inception v3: 87% -VGG19: 87% -ResNet50: 82% -Xception: 92% Sensitivity 84% and Specificity 97% Sensitivity 91% and Specificity 86%
Deep learning US(II) Xception +Top layer, 2 GPU minibatch20, 50 epoch Images from Park Surgical Clinic -training: 54,748(malignant(Invasive+ DCIS): 6,964) benign( FA, FCD, ADH, Normal) -validation13,600(20%) Accuracy: 92% Sensitivity: 68% Specificity: 95%
Quantitative shear wave ultrasound elastography: initial experience in solid breast masses Using the Aixplorer ultrasound system (SuperSonic Imagine, Aix en Provence, France), 53 solid breast lesions were identified in 52 consecutive patients. Mean elasticity cut off value of 50 kilopascals (kpa) was selected for benign/malignant differentiation. Shear wave elastography versus greyscale BI-RADS performance figures were sensitivity: 97% vs 87%, specificity: 83% vs 78%, positive predictive value (PPV): 88% vs 84%, negative predictive value (NPV): 95% vs 82% and accuracy: 91% vs 83% respectively. These differences were not statistically significant. Evans et al. Breast Cancer Research 2010, 12:R104
SonixEmbrace U system Somo-V Siemens ABVS GE invenia Techniscan Warmbath
Hitachi Sofia Delphinus SoftVue Reflection Attenuation Transmission Caperay US & x-ray 고해상도고조도상화술
MammouS
MammouS CAD -Segmentation
MRI Cat 4 lesion : Korean Health Insurance reimbuse. Multiple cystic and mass in both breast Cancer patients.
6 mm in MRI and Second look US Path: 3 mm DCIS with Bexcore
Breast Biopsy Options Invasive Open Surgical Breast Biopsy Minimally Invasive FNAB Breast cytology Core Needle Breast Biopsy Vacuum Assisted Breast Biopsy
FNAB Core Biopsy Gun Biopsy
Rebiopsy(surgical excision) after Percutaneous Biopsy - ADH - Phyllodes tumor - Discordance between imaging and histologic findings - Inadequate specimens - Controversy in papillary lesion radial scar atypical lobular hyperplasia LCIS - Larger volume of tissue or more contiguous sampling : lower the rebiopsy rate 14G core biopsy: 15 % 11G Mammotome: 9 % Philpotts LE, AJR 1999
Shortcomings of standard automated core biopsy 29% of breast lesions are heterogeneous yielding different histologic results from targeted center and periphery, thus sampling only part of a heterogeneous mass result in a misdiagnosis Morris et al Breast J 2002 Core biopsy findings of ADH underestimate the diagnosis of malignancy by 18 88% Joshi et al Breast J 2001
Breast related instruments and equipment
Thank You. Park Surgical Clinic Medical Park Ltd.