Silvio R. Pires, PhD, University Federal of São Paulo; Regina B. Medeiros, PhD; Simone Elias, PhD; Ana C. Patrocinio, PhD

Compare the performance between radiologists and residents to detect signals present in the phantom images from mamographic equipment and analyzed through viewbox and digital format.

The radiologists must follow the technology evolution and acquire skills that can be used as a resource for detecting signals in the image.[1] In this transition phase between analog and digital, the recognition of the new standard of image must be quickly absorbed, so the potential benefits of mammographic screening is kept.[2] So, a methodology is proposed in which a computational tool is utilized to aid in the complex multifatorial process of learning about these professionals, whether experienced or beginner.

A pilot project of training for detecting signals in images was initiated in 2006, and used printed film images (readings on viewboxes) and digital images (readings on monitors). The training accomplished theoretical activities with themes like digital systems qualification, image pre-processing, and others, as well as being based on phantom image readings at Medical Image Qualification Laboratory – QualIM.

A team of specialists and residents completed 1,920 readings out of a 60 images group taken from a statistical simulator (phantom ALVIM) generated in a same mammographic equipment.[3] Half of the images were generated adding 2 cm acrylic over the phantom in such way as to simulate a thick or big breast. Therefore, the phantom images were generated to represent a breast with normal thickness (4.5 cm) or higher (6.5 cm). All of the images were digitized by means of specific scanner for radiologic films (Lumiscan 75 – Lumisys) and displayed on the monitor. The readings were made on a specific viewbox for mamography (luminance between 3,000 and 3,500 cd/m2) and high resolution monitors (≥ 3 megapixels and luminance ≥ 500 cd/m2).The images were read in a darkened room for both hard and soft copy readings. The reading environment was controled regarding to the luminosity (≤ 20 lux) and noises that could impair the concentration.

During the readings, professionals focused on detecting phantom structures (microcalcification and fibers with different sizes) present in the image. Each finding was classified by the professional in 5 levels of confidence:100% (the object is definitely there), 75% (the object is likely there), 50% (presence of the object is uncertain), 25% (presence of the object is unlikely) and 0% (there definitely is no object).

The training software developed made manipulation of the images the on the monitors possible, making digital tools available (contrast and brightness adjust, zooming and inversion).[4,5] This software allows automatic analysis of the reader’s responses and his performance based on the Signals Detection Theory, thru statistical parameters: probability of detectability (Pdet), kappa values (k), true positive results (VP), false positive results (FP) and area under the ROC curve for different structures of different sizes.[6]

The parameters’ results were used to compare the performance of the professionals on the soft and hard copy readings. For specialists and residents the following results were gotten for the readings made on viewbox (analog images), respectively: probability of detectability (Pdet) 0.824(12) and 0.792(21); kappa (k) value 0.653(10) and 0.564(12); true positives (TP) 71% and 66%; false positive (FP) 1% and 7%; area under the ROC curve 0.803(22) and 0.763(41).

The results for soft and hard copy readings were, respectively: Pdet 0.848(14) and 0.823(17); Kappa value 0.712(6) and 0.641(8); TP 70% and 70%; false positive (FP) 5% and 5%; area under ROC curve 0.809(9) and 0.809(12). The difference for fibers detection rate (0,6mm size) between specialist and residents for the readings made on viewbox and monitors were, respectively: 15% and 5%.

The experience and training are factors that influence the professionals` performance regarding the objects detection in the image.[1] Studies conclude that the interpretation of the images on monitors are well accepted, the adaptation of the specialist is fast, and the accuracy and time spent on the interpretation are comparable to the conventional ones, since the professional is well trained.[7,8]

The results showed that the resident’s detectability on training is lower than the specialists in both systems of reading, mainly for fibers. It can also be the observed that the detection and correct response was improved due to training on monitor with digital tools.

For the least experienced professional, the results point to a lower perception in the structures recognition in the image and uncertainties in the fiber detection in which the contrast of the image is relevant. This lower detection is due mainly to the lack of hability and training, making the recognition and differentiation of actual injury and artifacts worse. Professionals, independently of training levels, presented better performance in the detectability of the structures for readings on monitors, as noticed by other researchers.[7,8]

These data reveal a higher sensibility and specificity of the professionals on digital systems. It is important to highlight that the conclusions presented herein are limited to the readings of phantom images, and cannot be extended to mamographic image readings.

The performance among radiologists and residents, when detecting signals present on phantom images, analyzed on monitors, were superior to the ones analyzed on viewboxes. The training, applied according to the methodology, proved to be effective among both groups of professionals. This methodology is a useful tool for training perception and for detection of structures present on mamographic images.

1. Elmore JG, Wells CK, Howard DH. Does Diagnostic Accuracy in Mammography Depend on Radiologist’s Experience? J Womens Health. 1998;7(4):443-9.
2. Obenauer S, Hermann KP, Marten K, et al. Soft copy versus hard copy reading in digital mammography. Journal of Digital Imaging. 2003;16(4):341-344,
3. Alvim Statistical Fantoma “Instruction Manual” Models 07-650 / 07-750 / 18-209, Nuclear Associates.
4. Pires SR, Medeiros, RB. Influence of Monitor Characteristics on the Signals Detection Present in the Mammographic Phantom Image. Proceedings of SPIE. 2008;Vol.6917:136-146,
5. Pires SR. Software gerenciador de uma base de dados e de imagens mamográficas classificadas segundo índice de qualidade. Radiol Bras. 2003;36(6):394-396.
6. Metz CE. Receiver operating characteristic analysis: A tool for the quantitative evaluation of observer performance and imaging systems,” Am. Coll. Radiol. 2006;3:413–422.
7. Hemminger BH. Soft Copy Display Requirements for Digital Mammography. Journal of Digital Imaging. 2003;Vol.16(3):292-305.
8. Obenauer S, Hermann KP, Marten K, Nagel et al. Soft copy versus hard copy reading in digital mammography. Journal of Digital Imaging. 2003;16(4):341-344.